TWI780382B - Microcontroller updating system and method - Google Patents

Abstract

A microcontroller updating system is provided in the invention. The microcontroller updating system may comprise a network platform, a first electronic device and a microcontroller. The first electronic device receives a training audio data or a training image data, uses a neural network algorithm to generate a recognition model according to the training audio data or training image data, and generates a header file according to the parameters of the recognition model, and converts the header file into a bin file and transmits the bin file to the network platform. The microcontroller obtains the bin file and updates the parameters corresponding to the recognition model according to the bin file.

Description

Microcontroller update system and method

The embodiments of the present invention are mainly related to a micro-controller update technology, especially related to the micro-controller update technology of updating the parameters of the identification model in the micro-controller through a programming file.

In the current recognition model based on machine learning technology, whether it is a smart image recognition model or a voice keyword recognition model, it is necessary to first train the data through a remote host to obtain training parameters (for example: weight ) and partial weight (bias)). Weights and biases play a very important role in neural network systems.

However, when the developer on the client side needs to update the parameters of the identification model, if the developer on the client side does not have sufficient machine learning knowledge background, he will not be able to compile the data correctly and efficiently to generate the required training parameters. Therefore, how to provide the developer of the client side to update the parameters of the identification model in the microcontroller more correctly and efficiently will be a topic worth discussing.

In view of the above-mentioned problems in the prior art, embodiments of the present invention provide a system and method for updating a microcontroller.

According to an embodiment of the present invention, a microcontroller updating system is provided. The microcontroller update system may include a network platform, a first electronic device and a microcontroller. The first electronic device can receive a training voice data or a training image data, and generate a training voice data based on a recurrent neural network algorithm, or a convolutional neural network algorithm based on the training image data. Identify the model, and generate a header file according to the parameters included in the identification model, convert the header file into a burning file, and upload the burning file to the network platform. The microcontroller can obtain the above-mentioned burning file, and update the above-mentioned parameters corresponding to the above-mentioned identification model according to the above-mentioned burning file.

In some embodiments, the microcontroller update system further includes a programming device. The burning device can obtain the above-mentioned burning file, and burn the above-mentioned burning file in one of the flash memory of the above-mentioned microcontroller.

In some embodiments, the microcontroller update system further includes a secure digital card. The secure digital card can be used to store the above-mentioned burning files uploaded to the above-mentioned network platform. The above-mentioned microcontroller obtains the above-mentioned burning file through the above-mentioned secure digital card.

In some embodiments, the above-mentioned parameters include the number of recognition categories corresponding to the above-mentioned recognition model, model size, weight value and partial weight value.

According to an embodiment of the present invention, a method for updating a microcontroller is provided. The steps of the above microcontroller updating method include: receiving a training voice data or a training image data; using a recursive neural network algorithm, according to the training voice data, or using a convolutional neural network algorithm, according to the training Generate a recognition model from the image data; generate a header file according to the parameters included in the above recognition model; Converting the above-mentioned header file into a burning file; uploading the above-mentioned burning file to a network platform; and obtaining the above-mentioned burning file, and updating the above-mentioned parameters corresponding to the above-mentioned identification model according to the above-mentioned burning file.

For other additional features and advantages of the present invention, those skilled in the art can make some changes according to the microcontroller update system and method disclosed in the implementation method of this case without departing from the spirit and scope of the present invention. obtained with retouching.

100: Microcontroller update system

110: The first electronic device

111: Radio device

112: Image capture device

113: The first processor

120: Internet platform

130: the second electronic device

140: microcontroller

141: Flash memory

142: storage device

143: second processor

200: Burning device

300: Secure Digital Card

S310~S360: Steps

FIG. 1 is a block diagram of a microcontroller updating system 100 according to an embodiment of the present invention.

FIG. 2A shows a block diagram of an updated microcontroller 140 according to one embodiment of the present invention.

FIG. 2B shows a block diagram of an updated microcontroller 140 according to one embodiment of the present invention.

FIG. 3 is a flowchart of a method for updating a microcontroller according to an embodiment of the present invention.

What is described in this chapter is the best way to implement the present invention. The purpose is to illustrate the spirit of the present invention and not to limit the protection scope of the present invention. .

FIG. 1 is a block diagram of a microcontroller updating system 100 according to an embodiment of the present invention. As shown in Figure 1, the microcontroller update system 100 It includes a first electronic device 110 , a network platform 120 , a second electronic device 130 , and a microcontroller 140 . It should be noted that the block diagram shown in FIG. 1 is only for the convenience of describing the embodiment of the present invention, but the present invention is not limited to FIG. 1 .

In some embodiments of the present invention, the first electronic device 110 may be a desktop computer, a notebook computer or other electronic devices with computing and data processing functions. In some embodiments of the present invention, the first electronic device 110 may be located in a server of the cloud system or a remote server.

In some embodiments of the present invention, the second electronic device 130 may be a desktop computer, a notebook computer or other electronic devices with computing and data processing functions. In some embodiments of the present invention, the second electronic device 130 may be an electronic device with a programming function.

In the embodiment of the present invention, the first electronic device 110 may include a sound receiving device 111 , an image capturing device 112 and a first processor 113 , but the present invention is not limited thereto. The sound collecting device 111 can be a microphone, but the present invention is not limited thereto. The sound receiving device can be used to receive at least one training voice data, and transmit the data to the first processor 113 . In addition, in the embodiment of the present invention, the image capture device 112 may be composed of a charge coupled device (Charge Coupled Device; CCD) or a complementary metal-oxide semiconductor (Complementary Metal-Oxide Semiconductor; CMOS) sensor . The image capture device 112 can be used to receive at least one training image data, and send the data to the first processor 113 . According to an embodiment of the present invention, the first electronic device 110 may simultaneously or sequentially receive one or more training image data (for example: sequentially input a set of training image data with "0" to "9" patterns, but the present invention not limited thereto) and/or one or more training speech data (for example: input a set of training speech data having pronunciations from "zero" to "nine" sequentially, but the present invention is not limited thereto).

In an embodiment of the present invention, the microcontroller 140 may include a flash memory 141, a storage device 142, and a second processor 143, but the present invention is not limited thereto.

In an embodiment of the present invention, when the first processor 113 receives training voice data or training image data, the first processor 113 will extract a plurality of voice features in the training voice data, or extract a plurality of voice features in the training image data. The complex image features of . When the first processor 113 receives the training voice data, the first processor 113 will extract a plurality of voice features in the training voice data; when the first processor 113 receives the training image data, the first processor 113 will Extract a plurality of image features in the training image data, and when the first processor 113 receives the training voice data and the training image data, the first processor 113 will sequentially extract the voice features and images according to the order received feature.

According to an embodiment of the present invention, the first processor 113 uses the Mel-scale Frequency Cepstral Coefficients (MFCC) algorithm to extract the speech features contained in the training speech data, but the present invention does not This is the limit. In some embodiments, other algorithms for extracting speech features can be used according to the actual implementation of the system.

According to an embodiment of the present invention, the first processor 113 analyzes each pixel in the training image data to obtain image features included in the training image data. For example, the first processor 113 may obtain the pixel value of each point in the training image data, and regard each pixel value as an image feature, but the invention is not limited thereto. In some embodiments, other algorithms for extracting image features may be used according to the actual implementation of the system.

According to an embodiment of the present invention, when the first processor 113 retrieves the word After the voice features are extracted, the first processor 113 can generate a specific number of complex voice parameters according to the extracted voice features. Next, the first processor 113 executes a procedure of a neural network algorithm. The first processor 113 inputs the generated speech parameters into a recurrent neural network (RNN) algorithm, and executes the program of the recurrent neural network algorithm to generate a recognition model. According to an embodiment of the present invention, after the first processor 113 extracts the speech features, the first processor 113 will generate a specific number of speech parameters (for example: 250, but the present invention does not limited), and present the speech parameters in the form of a one-dimensional feature set. According to an embodiment of the present invention, the first processor 113 will input speech parameters and a training answer (for example: the pronunciation of the answer "zero", but the present invention is not limited thereto) into the recurrent neural network algorithm, and execute Program of the recurrent neural network algorithm to generate the identification model.

According to an embodiment of the present invention, after the first processor 113 extracts the image features, the first processor 113 can generate a specific number of complex image parameters according to the extracted image features. Next, the first processor 113 executes a program of a convolutional neural network (CNN) algorithm. The first processor 113 will input the generated image parameters into a convolutional neural network algorithm, and execute the program of the convolutional neural network algorithm to generate a recognition model. According to an embodiment of the present invention, after the first processor 113 extracts the image features, the first processor 113 will generate a specific number of image parameters (for example: 250, but the present invention does not limited), and present the image parameters as a one-dimensional feature set. According to an embodiment of the present invention, the first processor 113 inputs image parameters and training answers (for example, images whose answer is "0", but the present invention is not limited thereto) into the deep neural network to generate a recognition model.

The neural network algorithm (for example: recurrent neural network algorithm and convolutional neural network algorithm) described in the embodiments of the present invention may also be other suitable deep learning or machine learning methods. Since the neural network algorithm is a known technology, it will not be repeated here.

According to an embodiment of the present invention, after the first processor 113 generates the identification model, the first processor 113 generates a header file according to the parameters included in the identification model. According to an embodiment of the present invention, the parameters included in the recognition model may include the number of recognition categories corresponding to the recognition model, model size, weight value and partial weight value, but the present invention is not limited thereto.

After generating the header file, the first processor 113 converts the header file into a burning file (for example: bin file), and uploads the burning file to the network platform 120 . When the client needs to update the microcontroller 140 , the client can obtain the burning file from the network platform 120 through the second electronic device 130 and use the burning file to update the microcontroller 140 . The following will be illustrated with Figures 2A-2B.

FIG. 2A shows a block diagram of an updated microcontroller 140 according to one embodiment of the present invention. As shown in FIG. 2A , the second electronic device 130 may be a burning device 200 . According to an embodiment of the present invention, when the microcontroller 140 is to be updated on the client side, the programming device 200 will obtain the programming file from the network platform 120, and use an online programming method, for example: in-system programming ( ISP, In System Programming), burn the programming file in one of the flash memory 141 of the microcontroller 140 to update the microcontroller 140. The second processor 143 of the microcontroller 140 can obtain the programming file from the flash memory 141 , and update the parameters of the identification model stored in the storage device 142 according to the information contained in the programming file.

Fig. 2B shows the modification according to another embodiment of the present invention. Block diagram of the new microcontroller 140 . As shown in FIG. 2B, according to an embodiment of the present invention, when the microcontroller 140 is to be updated on the client side, the second electronic device 130 can obtain the programming file from the network platform 120 and store the programming file in a Secure Digital (Secure Digital, SD) card 300 . Next, the user of the client can insert the SD card 300 into the SD card slot of the microcontroller 140 to update the microcontroller 140 . The second processor 143 of the microcontroller 140 can obtain the burning file from the SD card, and update the parameters of the identification model stored in the storage device 142 according to the information contained in the burning file.

FIG. 3 is a flowchart of a method for updating a microcontroller according to an embodiment of the present invention. The microcontroller update method is applicable to the microcontroller update system 100 . In step S310, the first electronic device of the microcontroller updating system 100 receives a training voice data or a training video data. In step S320, the first electronic device of the microcontroller updating system 100 generates a training voice data based on a recursive neural network algorithm, or a convolutional neural network algorithm based on the training image data. Identify the model. In step S330, the first electronic device of the microcontroller updating system 100 generates a header file according to the parameters included in the identification model. In step S340, the first electronic device of the microcontroller updating system 100 converts the header file into a burning file. In step S350, the first electronic device of the microcontroller updating system 100 uploads the above-mentioned burning file to a network platform. In step S360, the microcontroller of the microcontroller updating system 100 obtains the above-mentioned burning file, and updates the parameters corresponding to the above-mentioned identification model according to the above-mentioned burning file.

According to some embodiments of the present invention, the step S360 of the method for updating the microcontroller further includes, using a burning device to obtain the burning file from the network platform, and burning the burning file in a flash memory of the microcontroller middle. The microcontroller can update the parameters of the corresponding identification model according to the programming file.

According to some embodiments of the present invention, the step S360 of the method for updating the microcontroller further includes, using a secure digital (SD) card to store the burning file uploaded to the network platform. After the microcontroller obtains the burning file from the SD card, the microcontroller can update the parameters of the corresponding identification model according to the burning file.

According to some embodiments of the present invention, the parameters included in the recognition model include the number of recognition categories corresponding to the recognition model, model size, weight value, and partial weight value, but the present invention is not limited thereto.

According to the micro-controller update method proposed by the embodiment of the present invention, when the developer of the client needs to update the parameters of the identification model, the developer of the client only needs to obtain the burning file from the network platform to update the micro-controller. Parameters of the identification model in the controller. Therefore, the developer of the client can update the parameters of the identification model in the microcontroller more correctly and efficiently.

The serial numbers in this specification and the claims, such as "first", "second", etc., are only for convenience of description, and there is no sequential relationship between them.

The steps of the methods and algorithms disclosed in the description of the present invention can be directly applied to hardware and software modules or a combination of the two by executing a processor. A software module (including execution instructions and associated data) and other data can be stored in data memory, such as random access memory (RAM), flash memory (flash memory), read only memory (ROM) , Erasable Programmable Read-Only Memory (EPROM), Electronically Erasable Programmable Read-Only Memory (EEPROM), Temporary Register, Hard Disk, Portable Application Disk, CD-ROM ROM), DVD, or any other computer-readable storage medium format within the skill of the art. A storage medium can be coupled to a machine device, such as a computer/processor (for the convenience of description, in this specification Represented by a processor), the processor can read information (such as code) and write information to the storage medium. A storage medium can integrate a processor. An application specific integrated circuit (ASIC) includes a processor and storage media. A user equipment includes an ASIC. In other words, the processor and the storage medium are included in the user equipment without being directly connected to the user equipment. Furthermore, in some embodiments, any product suitable for a computer program includes a readable storage medium that includes code associated with one or more disclosed embodiments. In some embodiments, the product of the computer program may include packaging materials.

The above paragraphs use various levels of description. Obviously, the teachings herein can be implemented in many ways, and any specific structure or function disclosed in the examples is only a representative situation. According to the teaching of this article, any person familiar with the art should understand that each aspect disclosed in this article can be implemented independently or two or more aspects can be implemented in combination.

Although the disclosure has been disclosed above with the embodiment, it is not intended to limit the disclosure. Anyone who is familiar with this technology can make some changes and modifications without departing from the spirit and scope of the disclosure. Therefore, the protection scope of the invention The ones defined in the scope of the attached patent application shall prevail.

S310~S360: Steps

Claims (8)

A microcontroller update system, comprising: a network platform; a first electronic device, receiving a training voice data or a training image data, by a recursive neural network algorithm, according to the training voice data, or by a The convolutional neural network algorithm generates a recognition model based on the above-mentioned training image data, and generates a header file according to the parameters contained in the above-mentioned recognition model, and converts the above-mentioned header file into a burning file, and burns the above-mentioned The recording file is uploaded to the above-mentioned network platform; and a microcontroller, which obtains the above-mentioned burning file, and updates the above-mentioned parameters corresponding to the above-mentioned identification model according to the above-mentioned burning file. The microcontroller update system described in item 1 of the scope of the patent application further includes: a burning device, which obtains the above-mentioned burning file, and burns the above-mentioned burning file into a flash memory of the above-mentioned microcontroller . The microcontroller update system described in Item 1 of the scope of the patent application further includes: a secure digital card for storing the above-mentioned burning file uploaded to the above-mentioned network platform, wherein the above-mentioned microcontroller obtains the above-mentioned burning file through the above-mentioned secure digital card record. The microcontroller updating system described in item 1 of the scope of the patent application, wherein the above-mentioned parameters include the number of recognition categories, model size, weight value and partial weight value corresponding to the above-mentioned recognition model. A microcontroller updating method, comprising: receiving a training voice data or a training image data; A recurrent neural network algorithm is used to generate a recognition model based on the above-mentioned training voice data, or a convolutional neural network algorithm is used to generate a recognition model based on the above-mentioned training image data; a header is generated according to the parameters contained in the above-mentioned recognition model converting the above-mentioned header file into a burning file; uploading the above-mentioned burning file to a network platform; and obtaining the above-mentioned burning file, and updating the above-mentioned parameters corresponding to the above-mentioned identification model according to the above-mentioned burning file. The microcontroller update method described in item 5 of the scope of the patent application further includes: obtaining the above-mentioned burning file from the above-mentioned network platform by a burning device; burning the above-mentioned burning file in one of the microcontrollers in the flash memory; and update the above-mentioned microcontroller according to the above-mentioned burning file. The microcontroller update method described in item 5 of the scope of the patent application further includes: using a secure digital card to store the above-mentioned burning file uploaded to the above-mentioned network platform; updating the above-mentioned burning file obtained according to the above-mentioned secure digital card the aforementioned microcontroller. The microcontroller update method described in item 5 of the patent application, wherein the above-mentioned parameters include the number of recognition categories corresponding to the above-mentioned recognition model, model size, weight value and partial weight value.

Images