TWI788651B - Control system for a touch device and method thereof - Google Patents

Abstract

A control system for a touch device which has a touch sensor includes a sensing circuit for sensing the touch sensor to generate a plurality of sensing values, a processor for generating a sensing picture according to the plurality of sensing values, and a convolutional neural network (CNN) for processing the sensing picture so as to generate a characteristic information and generate an identification information according to the characteristic information. The identification information is used to determine a state of the touch sensor.

Description

Control system and method for touch device

The present invention relates to a control system and method for a touch device, in particular to a control system with a convolutional neural network (CNN) and a method thereof.

In the touch device, a judging mechanism is needed to identify the type of the object touching the touch device or the state of the touch device. 1 to 4 show sensing images generated by the touch device in response to different objects or states, and the sensing images include multiple sensing quantities corresponding to different positions of the touch device. Figure 1 shows the sensing image when the object is a water droplet, Figure 2 shows the sensing image when the object is a Near-Field Communication (NFC) card, Figure 3 shows the sensing image when the object is a floating metal object Fig. 4 shows the sensing image when the object is a finger and the touch device is in an abnormal state with noise. The touch device needs to judge the type of the object touching the touch device or the state of the touch device according to the sensed sensing image to perform corresponding operations. When ignoring the operation of the object, or judging that the touch device is in an abnormal state, re-execute the calibration procedure.

The purpose of the present invention is to provide a control system and method for a touch device using a convolutional neural network.

According to the present invention, a control system for a touch device includes: a sensing circuit, A touch sensor for sensing the touch device to generate a plurality of sensing values; a processor connected to the sensing circuit to generate a sensing image according to the plurality of sensing values; and a convolutional neural network , for processing the sensing image to generate feature information, and generating identification information according to the feature information; wherein the processor judges the state of the touch sensor according to the identification information.

According to the present invention, a control system for a touch device includes: a sensing circuit for sensing a touch sensor of the touch device to generate a plurality of sensing values; a processor connected to the sensing circuit , generating a sensing image according to the plurality of sensing quantities, and performing object segmentation processing on the sensing image to determine a sub-image; and a convolutional neural network, used to process the sub-image to generate feature information , and generate identification information according to the characteristic information; wherein the processor judges an object type according to the identification information.

According to the present invention, a control system for a touch device includes: a sensing circuit for sensing a touch sensor of the touch device to generate a plurality of sensing values; a processor connected to the sensing circuit , generating a sensing image according to the plurality of sensing quantities; a convolutional neural network for processing the sensing image to generate feature information, and generating identification information according to the feature information; and a host connected to the processing The touch sensor receives the sensing image, and judges the state of the touch sensor according to the identification information.

According to the present invention, a control system for a touch device includes: a sensing circuit for sensing a touch sensor of the touch device to generate a plurality of sensing values; a processor connected to the sensing circuit , generating a sensing image according to the plurality of sensing quantities; a convolutional neural network for processing a sub-image to generate feature information, and generating identification information according to the feature information; and a host connected to the processing The device judges the type of the object according to the identification information; wherein the host or the processor performs object segmentation processing on the sensed image to generate the sub-image.

According to the present invention, a method for a touch control device includes the following steps: obtaining a sensing image of a touch sensor of the touch control device, the sensing image including a plurality of sensing values; The neural network processes the sensing image to generate feature information and generates identification information according to the feature information; and judges the state of the touch sensor according to the identification information.

According to the present invention, a method for a touch device includes the following steps: obtaining a sensing image of a touch sensor of the touch device, where the sensing image includes a plurality of sensing quantities; Segmentation processing is used to determine a sub-image; a convolutional neural network is used to process the sub-image to generate feature information, and identification information is generated according to the feature information; and an object type is judged according to the identification information.

The present invention recognizes the type of a contact object or the state of a touch device through a convolutional neural network, and has the advantages of fast efficiency, convenience and high accuracy.

20: Touch device

22:Touch sensor

222: Sensing point

24: Controller

24A: Controller

24B: Controller

242: Sensing circuit

243: memory

244: Processor

2442: Convolutional Neural Network Programs

245: Processor

245B: Processor

246: memory

247: Convolutional Neural Network Circuits

248: memory

249: memory

26: Host

262: Convolutional Neural Network Program

30: Feature extraction part

32: Classification section

34: Image

40: memory

42: Memory

44: Convolutional Neural Network Circuit

Figure 1 shows the sensing image when the object is a water drop.

Figure 2 shows the sensing image when the object is an NFC card.

Figure 3 shows the sensing image when the object is a floating metal object.

FIG. 4 shows a sensing image when the object is a finger and the touch device is in an abnormal state with noise.

Figure 5 shows a first embodiment of the control system of the present invention.

Figure 6 shows a second embodiment of the control system of the present invention.

Figure 7 shows the basic architecture of a convolutional neural network.

FIG. 8 shows the method for identifying the state of the touch sensor of the present invention.

FIG. 9 shows that the present invention recognizes the types of objects that touch or approach the touch sensor.

Fig. 10 shows a third embodiment of the control system of the present invention.

Fig. 11 shows a fourth embodiment of the control system of the present invention.

Figure 5 illustrates a first embodiment of the control system of the present invention. In FIG. 5, the touch device 20 It includes a touch sensor 22 and a control system, and the control system includes a controller 24 and a host 26 . In one embodiment, the touch sensor 22 is a capacitive touch sensor, which has a plurality of electrodes TX1 - TX4 and RX1 - RX4 , and the intersections between the electrodes form the sensing point 222 . The layout of the electrodes TX1 - TX4 and RX1 - RX4 in FIG. 5 is just an example of the touch sensor 22 , but the present invention is not limited thereto. The controller 24 includes a sensing circuit 242 , a processor 244 , a memory 246 and a memory 248 . The sensing circuit 242 is connected to the touch sensor 22 and is used for sensing capacitance values of a plurality of sensing points 222 on the touch sensor 22 to generate a plurality of sensing values dV. The processor 244 generates a sensing image SP according to a plurality of sensing values dV from the sensing circuit 242 , as shown in FIG. 1 to FIG. 4 . The processor 244 is connected to the memories 246 and 248 . The processor 244 implements the convolutional neural network program 2442 with firmware, and the convolutional neural network program 2442 has the ability of inference. The memory 246 stores the parameters Dp required for the operation of the convolutional neural network program 2442. The memory 246 can be a ROM (Read-Only Memory), or a RAM (Random Access Memory) preloaded with initial values. 246 is not limited to ROM and RAM. The parameter Dp is pre-generated by the convolutional neural network training program with the same structure as the convolutional neural network program 2442 on the computer. When the convolutional neural network program 2442 performs different recognition functions, the required parameter Dp is also different. The memory 248 is connected to the processor 244 for storing temporary information or data generated by the convolutional neural network program 2442 of the processor 244 during operation. The memory 248 may be but not limited to RAM. In one embodiment, the memories 246 and 248 can also be combined into one memory. The host 26 may be a central processing unit (CPU) of an electronic device, such as a CPU of a notebook computer, or an embedded controller (Embedded Controller; EC), or a keyboard controller (KeyBoard Controller; KBC).

Figure 6 illustrates a second embodiment of the control system of the present invention. The architecture of the controller 24 of FIG. 5 is substantially the same as that of the controller 24A of FIG. 6 . In the controller 24A of FIG. 6 , the sensing circuit 242 is connected to the touch sensor 22 for sensing a plurality of sensing points 222 of the touch sensor 22 to generate a plurality of sensing values dV. The processor 245 generates a sensing image SP according to a plurality of sensing values dV from the sensing circuit 242 . The convolutional neural network circuit 247 is connected to the processor 245 , the memories 243 and 249 . The memory 243 stores the convolutional neural network circuit 247 The parameter Dp required for the operation of the memory 243 can be, but not limited to, a ROM or a RAM pre-loaded with initial values. The parameter Dp is pre-generated by the convolutional neural network training program on the computer with the same structure as the convolutional neural network circuit 247 . When the convolutional neural network circuit 247 performs different recognition functions, the required parameter Dp is also different. The memory 249 is connected to the convolutional neural network circuit 247 for storing temporary information or data generated by the convolutional neural network circuit 247 during operation. The memory 247 may be but not limited to RAM. In an embodiment, the memories 243 and 249 can also be combined into one memory.

The present invention utilizes a convolutional neural network to determine the state of the touch sensor 22 or the type of an object touching (or approaching) the touch sensor 22 . The convolutional neural network program 2442 in FIG. 5 implements the convolutional neural network with firmware. The convolutional neural network circuit 247 in FIG. 6 implements the convolutional neural network in the form of a hardware circuit. Both the convolutional neural network program 2442 and/or the convolutional neural network circuit 247 have the basic structure of the convolutional neural network shown in FIG. 7 , which can be divided into a feature extraction part 30 and a classification part 32 . The feature extraction part 30 is used for convolution operation and subsampling operation. The main function of the convolution operation is feature extraction, and the main function of the subsampling operation is to reduce the amount of image data and retain important information. The classification part 32 performs classification according to the retrieved feature information. As shown in FIG. 7 , when an image 34 including a number 3 is input into the convolutional neural network, the feature extraction part 30 will extract features of the image 34 to generate a feature information DF. The feature information DF is provided to the classification part 32 for classification to generate identification information DI, and the identification information DI is used to determine that the number in the image 34 is 3. Taking the recognition of the number 3 as an example, in the process of training the convolutional neural network to recognize the number 3, it is necessary to provide various images of the number 3 to the convolutional neural network. After the convolutional neural network extracts the feature information of the image, Put it into a numeric feature group. The convolutional neural network can recognize the number 3 in the image based on the feature information in these number feature groups. The convolutional neural network is a fairly mature technology, so its details will not be described in detail. The convolutional neural network used in the present invention may be a standard convolutional neural network architecture, or an extended and deformed architecture of the convolutional neural network. The feature extraction part 30 and the classification part 32 can be realized by firmware or hardware circuits.

The methods shown in FIG. 8 and FIG. 9 are described below with the control system shown in FIG. 5 .

FIG. 8 is a method for identifying the state of a touch sensor. Steps S10 and S11 are to obtain the sensing image of the touch sensor 22 of the touch device 20 . In step S10 , the sensing circuit 242 of the controller 24 senses the touch sensor 22 to generate a plurality of sensing values dV. Next, the processor 244 generates a sensing image SP according to a plurality of sensing values dV provided by the sensing circuit 242 , as shown in step S11 . The sensing image SP includes sensing values dV of each sensing point 222 in the touch sensor 22 . After the sensing image SP is acquired, go to step S12.

Step S12 is to identify the state of the touch sensor 22 according to the sensing image SP by the convolutional neural network program 2442 . In step S12 , the convolutional neural network program 2442 processes the sensing image SP to generate feature information DF1 , and generates identification information DI1 according to the feature information DF1 . In step S14 , the processor 244 determines the state of the touch sensor 22 according to the identification information DI1 . For example, the content of the identification information DI1 generated by the convolutional neural network program 2442 includes that the probability that the touch sensor 22 has water is 10%, and the probability that the touch sensor 22 is disturbed by noise is 90%. Obviously, the probability of the touch sensor 22 being interfered by noise is high. Therefore, according to the identification information DI1 , the processor 244 can determine whether there is noise in the state of the touch sensor 22 . Next, the processor 244 can perform corresponding processing, such as limiting single-finger operations, or changing the frequency of the driving signal applied to the touch sensor 22 .

To implement the embodiment of FIG. 8 , a convolutional neural network training program CT1 with the same structure as the convolutional neural network program 2442 needs to be provided on the computer in advance, and the convolutional neural network training program CT1 is also implemented by a programming language. In order for the convolutional neural network program 2442 to recognize the states of various touch sensors, such as noise interference, floating contact, and water droplets, it is necessary to pre-train the convolutional neural network training program CT1 to obtain the convolutional neural network program 2442 parameters required for identification. Taking the training of the convolutional neural network training program CT1 to recognize the state of the touch sensor 22 being disturbed by noise as an example, the training process includes providing multiple noises to the touch sensor 22, each time providing the noise Not the same in location, intensity or range. The touch sensor 22 is disturbed by multiple noises, so that the processor 244 obtains various sensing images SP with different sensing quantity distributions. These sensing images SP are extracted by the convolutional neural network training program CT1 to generate convolution The neural network program 2442 obtains the parameter Dp required by the feature information DF1 and the identification information DI1. The obtained parameter Dp is stored in the memory 246 for use when the convolutional neural network program 2442 recognizes the state of the touch sensor 22 . In this way, the convolutional neural network program 2442 has the ability to identify whether there is noise in the state of the touch sensor 22 . The convolutional neural network training program CT1 can also be trained to recognize other states of the touch sensor 22 , so that the convolutional neural network program 2442 is also capable of recognizing more different states of the touch sensor 22 . The process in the middle is roughly the same, and will not be repeated here.

The method shown in FIG. 9 is used to identify the type of object touching or approaching the touch sensor 22 . Steps S10 and S11 are the same as those in FIG. 8 . In step S16 , the processor 244 performs object segmentation processing on the sensed image SP to determine at least one sub-image. The object segmentation process determines one or more object regions from the sensing image SP, and then determines a sub-image according to each object region, and the sub-image includes the image of the object region. In other words, the sub-image is a part of the sensing image SP, which includes a plurality of sensing quantities. For example, after the sensing image SP generated in response to two objects touching the touch sensor 22 is subjected to object segmentation processing, the processor 244 defines two object regions from the sensing image SP, and the processor 244 defines two object regions according to the two objects. The object regions respectively determine two corresponding sub-images, and each sub-image includes an image of the object region.

Step S17 is to use the convolutional neural network program 2442 to identify the type of the object touching or approaching the touch sensor 22 according to the sub-image determined in step S16. In step S17, the convolutional neural network program 2442 processes the sub-image to generate feature information DF2, and generates identification information DI2 according to the feature information DF2. If there are two sub-images, the convolutional neural network program 2442 needs to process these two sub-images to generate two pieces of feature information DF2 and two pieces of identification information DI2. In step S18, the processor 244 determines an object type according to each identification information DI2. For example, the content of the identification information DI2 includes that the probability of the type of the object is water is 90%, the probability of the finger is 7%, and the probability of the stylus is 3%. Obviously, the probability of the object being water is particularly high. Therefore, according to the identification information DI2, the processor 244 can determine that the object on the touch sensor 22 is water. Similarly, if there are two pieces of identification information DI2, the processor 244 The type of the object will be judged respectively according to each piece of identification information DI2. Next, the processor 244 can perform corresponding processing. For example, when the contact object is water, the processor 244 does not calculate and output the coordinates, and when the contact object is a stylus, adjust the gain of the sensing circuit 242 (Gain).

To implement the embodiment of FIG. 9 , the convolutional neural network program 2442 needs to be trained in advance to recognize various types of objects, such as fingers, water droplets, and stylus. Taking the training of the convolutional neural network program 2442 to understand water as an example, the training process includes dropping water droplets of various sizes on the touch sensor 22 for many times, and the positions and shapes of the water droplets are different each time. Dropping water on the touch sensor 22 multiple times enables the processor 244 to obtain various sensing images SP with different sensing quantity distributions. These sensing images SP are extracted by the convolutional neural network training program CT2 to generate the parameter Dp required by the convolutional neural network program 2442 to obtain the feature information DF2 and the identification information DI2. The obtained parameter Dp is stored in the memory 246 for the convolutional neural network program 2442 to identify the object species, so that the convolutional neural network program 2442 has the ability to identify water. The convolutional neural network training program CT2 can also be trained to recognize other types of objects, so that the convolutional neural network program 2442 also has the ability to recognize more different objects. The process in the middle is roughly the same, and will not be repeated here.

The methods shown in FIG. 8 and FIG. 9 are also applicable to the architecture of FIG. 6 . Step S12 can also be implemented by the processor 245 controlling the operation of the convolutional neural network circuit 247 . Therefore, step S12 should be understood as processing the sensing image SP through a convolutional neural network to generate feature information DF1 and generating identification information DI1 according to the feature information DF1 . Step S17 can also be implemented by the processor 245 controlling the operation of the convolutional neural network circuit 247 . Therefore, step S17 should be understood as processing a sub-image by a convolutional neural network to generate feature information DF2, and generating identification information DI2 according to the feature information DF2.

In one embodiment, after the sensing image SP is generated in step S11, the processor 244 (or 245) performs preprocessing on the sensing image SP first. The preprocessing includes but is not limited to dealing with noise or compensating for abnormal values. Then, step S12 or S16 is performed by using the preprocessed sensing image SP.

In one embodiment, the control system 24 of FIG. 5 and the control system 24A of FIG. 6 may be an integrated circuit device.

According to the present invention, as long as sufficient sensing images SP are provided in advance to train the convolutional neural network, and the required parameters are pre-stored in the memory (246 or 243), the controller 24 (or 24A) can learn to use The image identifies the type of contact object or the state of the touch sensor. Therefore, the present invention has the advantages of simplicity and high recognition accuracy.

Fig. 10 shows a third embodiment of the control system of the present invention. The control system in FIG. 10 includes a controller 24B, a host 26 , a memory 40 and a memory 42 . The controller 24B has a sensing circuit 242 and a processor 245B. The controller 24B can be an integrated circuit device. The sensing circuit 242 is connected to the touch sensor 22 and is used for sensing capacitance values of a plurality of sensing points 222 on the touch sensor 22 to generate a plurality of sensing values dV. The processor 245B generates a sensing image SP according to a plurality of sensing values dV from the sensing circuit 242 . The host 26 is connected to the processor 245B, the memory 40 and the memory 42 . The host 26 has a convolutional neural network program 262 implemented in firmware. The memory 40 is connected to the host computer 26 for storing the parameters Dp required for the operation of the convolutional neural network program 262. The memory 40 can be but not limited to ROM or RAM pre-loaded with initial values. The parameter Dp is pre-generated by the convolutional neural network training program with the same structure as the convolutional neural network program 262 on the computer. When the convolutional neural network program 262 performs different recognition functions, the required parameter Dp is also different. The memory 42 is connected to the host computer 26 for storing temporary information or data generated by the convolutional neural network program 262 during operation. The memory 42 may be but not limited to RAM. In an embodiment, the memories 40 and 42 can also be combined into one memory. In one embodiment, the memory 40 may be a ROM or a flash memory in the host 26 , and the memory 42 may be a RAM in the host 26 . The host 26 may be the CPU, EC or KBC of the electronic device.

Figure 11 illustrates a fourth embodiment of the control system of the present invention. The control system in FIG. 11 includes the controller 24B, the host computer 26 , the memory 40 and the memory 42 as in FIG. 10 . The difference is that the control system in FIG. 11 also includes a convolutional neural network circuit 44 . The convolutional neural network circuit 44 implements a convolutional neural network in the form of a hardware circuit. The convolutional neural network circuit 44 is connected to the host computer 26 and the memories 40 and 42 . The memory 40 stores the parameters Dp required by the operation of the convolutional neural network circuit 44 . The parameter Dp is obtained via the computer in advance Above, generated by the convolutional neural network training program with the same structure as the convolutional neural network circuit 44. When the convolutional neural network circuit 44 performs different recognition functions, the required parameters Dp are also different. The memory 42 is connected to the convolutional neural network circuit 44 for storing temporary information or data generated by the convolutional neural network circuit 44 during operation. In one embodiment, the convolutional neural network circuit 44 may be integrated into the host 26 .

The convolutional neural network formula 262 in FIG. 10 and the convolutional neural network circuit 44 in FIG. 11 are similar to the convolutional neural network formula 2442 in FIG. 5 and the convolutional neural network circuit 247 in FIG. 6 , respectively. Refer to FIG. 7 for the basic architecture of the convolutional neural network program 262 and the convolutional neural network circuit 44 .

The methods shown in FIG. 8 and FIG. 9 are also applicable to the architecture of FIG. 10 . Referring to FIG. 8 and FIG. 10 , in step S10 , the sensing circuit 242 of the controller 24B senses the touch sensor 22 to generate a plurality of sensing values dV. Next, the processor 245B generates a sensing image SP according to a plurality of sensing values dV provided by the sensing circuit 242 , as shown in step S11 . The sensing image SP includes sensing values dV of each sensing point 222 in the touch sensor 22 . After obtaining the sensing image SP, the processor 245B transmits the sensing image SP to the host 26 to perform step S12.

Step S12 is to identify the state of the touch sensor 22 according to the sensing image SP by the convolutional neural network program 262 . In step S12 , the convolutional neural network program 262 of the host computer 26 processes the sensing image SP to generate feature information DF1 , and generates identification information DI1 according to the feature information DF1 . In step S14 , the host 26 judges the state of the touch sensor 22 according to the identification information DI1 . The method of judging the state and the training method of the convolutional neural network program 262 are the same as those of the convolutional neural network program 2442 in FIG. 5 , so details are not repeated here.

After determining the state of the touch sensor 22 in step S14 , the host 26 can notify the controller 24B of the determined state, so that the controller 24B can perform corresponding processing according to the state of the touch sensor 22 . For example, when it is determined that there is water or noise on the touch sensor 22, the controller 24B can adjust the parameters used to process the sensed image SP, or send instructions to the sensing circuit 242 to change the sensitivity to the touch sensing. The scanning mode or scanning frequency of the device 22. Scanning methods include but are not limited to self-capacity scanning and mutual-capacity scanning.

Referring to FIG. 9 and FIG. 10 , in step S10 , the sensing circuit 242 of the controller 24B senses the touch sensor 22 to generate a plurality of sensing values dV. Next, the processor 245B generates a sensing image SP according to a plurality of sensing values dV provided by the sensing circuit 242 , as shown in step S11 . The sensing image SP includes sensing values dV of each sensing point 222 in the touch sensor 22 . In step S16 , after obtaining the sensed image SP, the processor 245B performs object segmentation processing on the sensed image SP to determine at least one sub-image, and then the processor 245B transmits the at least one sub-image to the host 26 . In an embodiment, step S16 may also be that after the processor 245B transmits the sensing image SP to the host 26, the host 26 performs object segmentation processing on the sensing image SP to determine at least one sub-image. The operation and principle of the object segmentation processing are as described above, and will not be repeated here.

Step S17 is to use the convolutional neural network program 262 to identify the type of the object touching or approaching the touch sensor 22 according to the sub-image determined in step S16. In step S17, the convolutional neural network program 262 processes the sub-image to generate feature information DF2, and generates identification information DI2 according to the feature information DF2. If there are two sub-images, the convolutional neural network program 262 needs to process these two sub-images to generate two pieces of feature information DF2 and two pieces of identification information DI2. In step S18, the host 26 determines an object type according to each identification information DI2. The manner of judging the type of the object by the convolutional neural network program 262 and the training method are the same as those of the convolutional neural network program 2442 in FIG. 5 , so details are not repeated here.

The methods shown in FIG. 8 and FIG. 9 are also applicable to the architecture of FIG. 11 . Step S12 can also be implemented by the host 26 controlling the operation of the convolutional neural network circuit 44 . Therefore, step S12 should be understood as processing the sensing image SP through a convolutional neural network to generate feature information DF1 and generating identification information DI1 according to the feature information DF1 . Step S17 can also be implemented by the host 26 controlling the operation of the convolutional neural network circuit 44 . Therefore, step S17 should be understood as processing a sub-image by a convolutional neural network to generate feature information DF2, and generating identification information DI2 according to the feature information DF2.

In one embodiment, after the sensing image SP is generated in step S11, the processor 245B performs preprocessing on the sensing image SP first. The preprocessing includes but is not limited to dealing with noise or compensating for abnormal values. Then, step S12 or S16 is performed by using the preprocessed sensing image SP.

According to the present invention, as long as sufficient sensing images SP are provided in advance to train the convolutional neural network program 262 (or convolutional neural network circuit 44), the host computer 26 can learn to recognize the type of the contact object or touch object according to the sensing image. control sensor status. Therefore, the present invention has the advantages of simplicity and high recognition accuracy.

The above descriptions of the preferred embodiments of the present invention are for the purpose of illustration, and are not intended to limit the present invention to the disclosed form. It is possible to modify or change based on the above teachings or learning from the embodiments of the present invention. The embodiment is selected and described in order to explain the principle of the present invention and to allow those familiar with the art to use the present invention in various embodiments for practical application. Decide.

20: Touch device

22:Touch sensor

222: Sensing point

24: Controller

242: Sensing circuit

244: Processor

2442: Convolutional Neural Network Programs

246: memory

248: memory

26: Host

Claims (32)

A control system for a touch device, the touch device includes a touch sensor, the control system includes: a sensing circuit, used to sense the touch sensor to generate a plurality of inductive quantities; A processor, connected to the sensing circuit, generates a sensing image according to the plurality of sensing quantities; and a convolutional neural network, used to process the sensing image to generate feature information, and generate a recognition based on the feature information information; wherein the processor judges the state of the touch sensor as noise interference, floating or water drop according to the identification information. The control system of claim 1, wherein the convolutional neural network is implemented in firmware of the processor. The control system according to claim 1, wherein the convolutional neural network is implemented by a hardware circuit. The control system of claim 2 further includes a memory connected to the processor for storing parameters required for the operation of the convolutional neural network. The control system of claim 3 further includes a memory connected to the convolutional neural network for storing parameters required for the operation of the convolutional neural network. The control system according to claim 1, wherein the processor further includes preprocessing the sensing image, the preprocessing includes processing noise or compensating abnormal values, and the processor provides the preprocessed sensing image to the A convolutional neural network is used to generate the recognition information. A control system for a touch device, the touch device includes a touch sensor, the control system includes: a sensing circuit, used to sense the touch sensor to generate a plurality of inductive quantities; A processor, connected to the sensing circuit, generates a sensing image according to the plurality of sensing quantities, and performing object segmentation processing on the sensed image to determine a sub-image; and a convolutional neural network for processing the sub-image to generate feature information, and generate identification information based on the feature information; wherein the The processor determines an object type according to the identification information. The control system as claimed in claim 7, wherein the convolutional neural network is implemented in firmware of the processor. The control system as claimed in claim 7, wherein the convolutional neural network is implemented by a hardware circuit. The control system of claim 8 further includes a memory connected to the processor for storing parameters required for the operation of the convolutional neural network. The control system of claim 9 further includes a memory connected to the convolutional neural network for storing parameters required for the operation of the convolutional neural network. The control system according to claim 7, wherein the processor performs preprocessing on the sensed image before performing the object segmentation processing, and the preprocessing includes processing noise or compensating abnormal values. A control system for a touch device, the touch device includes a touch sensor, the control system includes: a sensing circuit, used to sense the touch sensor to generate a plurality of inductive quantities; A processor, connected to the sensing circuit, generates a sensing image according to the plurality of sensing quantities; a convolutional neural network is used to process the sensing image to generate feature information, and generate identification information according to the feature information and a host connected to the processor to receive the sensing image, and judge the state of the touch sensor as noise interference, floating or water drop according to the identification information. The control system according to claim 13, wherein the host is a central processing unit, an embedded controller or a keyboard controller. The control system according to claim 13, wherein the convolutional neural network is implemented with firmware of the host. The control system according to claim 13, wherein the convolutional neural network is implemented by a hardware circuit. The control system according to claim 16, wherein the convolutional neural network is integrated in the host. The control system according to claim 15 further includes a memory connected to the host for storing parameters required for the operation of the convolutional neural network. The control system of claim 16 further includes a memory connected to the convolutional neural network for storing parameters required for the operation of the convolutional neural network. The control system according to claim 13, wherein the processor further includes pre-processing the sensing image, the pre-processing includes processing noise or compensating abnormal values, and the processor provides the pre-processed sensing image to the A convolutional neural network is used to generate the recognition information. A control system for a touch device, the touch device includes a touch sensor, the control system includes: a sensing circuit, used to sense the touch sensor to generate a plurality of inductive quantities; A processor, connected to the sensing circuit, generates a sensing image according to the plurality of sensing quantities; a convolutional neural network is used to process a sub-image to generate feature information, and generate identification information according to the feature information and a host connected to the processor and judging the type of the object according to the identification information; wherein the host or the processor generates the sub-image after performing object segmentation processing on the sensed image. The control system according to claim 21, wherein the host computer is a central processing unit, an embedded controller or a keyboard controller. The control system as in claim 21, wherein the convolutional neural network is implemented with firmware of the host. The control system according to claim 21, wherein the convolutional neural network is implemented by a hardware circuit. The control system of claim 24, wherein the convolutional neural network is integrated in the host. The control system of claim 23 further includes a memory connected to the host for storing parameters required for the operation of the convolutional neural network. The control system of claim 24 further includes a memory connected to the convolutional neural network for storing parameters required for the operation of the convolutional neural network. The control system according to claim 21, wherein before the object segmentation process is performed on the sensed image, the processor firstly performs preprocessing on the sensed image, and the preprocessing includes processing noise or compensating abnormal values. A method for a touch control device, comprising the following steps: a. Obtaining a sensing image of a touch sensor of the touch control device, the sensing image including a plurality of sensing values; b. The network processes the sensing image to generate feature information and generate identification information according to the feature information; and c. judging the state of the touch sensor as noise interference, floating or water drop according to the identification information. The method according to claim 29, wherein before the step b, it further includes preprocessing the sensed image, and the preprocessing includes processing noise or compensating abnormal values. A method for a touch device, comprising the following steps: a. obtaining a sensing image of a touch sensor of the touch device, the sensing image including a plurality of Sensing amount; b. performing object segmentation processing on the sensing image to determine a sub-image; c. processing the sub-image by a convolutional neural network to generate a feature information, and generating a recognition information based on the feature information and d. judging an object type according to the identification information. The method according to claim 31, further comprising preprocessing the sensed image before step b, the preprocessing including processing noise or compensating abnormal values.

Images