Image classification method, system and terminal based on DCPI resin

Abstract

The application discloses a kind of based on DCPI resin's image classification method, system and terminal, the method includes: constructing based on DCPI resin's photonic neural network, the photonic neural network includes input layer, optical weight self-learning layer and output layer;The photonic neural network is trained, and target network is obtained;Target image is collected, and the target image is input to the input layer of the target network and is preprocessed, and the preprocessed image is obtained, and the preprocessed image is input to the optical weight self-learning layer of the target network and is inferred, and inference result is obtained, and inference result is input to the output layer of target network and is classified, and the classification result of target image is output.The photonic neural network based on DCPI resin in the application is an independent photonic neural network not dependent on artificial neural network, and the generation of weight does not require a large amount of computing resources and energy consumption, which is an important step of artificial intelligence.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photonic neural networks, and particularly relates to an image classification method and system based on DCPI resin, a terminal and a computer readable storage medium. BACKGROUND

[0002] With the vigorous development of integrated circuit technology, artificial intelligence has rapidly become one of the most active fields in modern society. However, due to the slowing down of Moore's law, the Von Neumann bottleneck of existing computer architecture, and the inherent defects that processing speed is proportional to energy loss, people once again turn their eyes to the field of optics. It is hoped that the neural network will be built in the field of optics, and the characteristics of parallel processing and high speed and low power consumption of light will be used to overcome various limitations faced by electronic neural networks, so as to promote new breakthroughs in artificial intelligence.

[0003] A photonic neural network is an artificial neural network that uses optical technology for information processing. It performs calculations through photons rather than electrons, has the ability of super parallel processing and transmission of information, and is suitable for high-density lead and direct processing of image scenes. The development of photonic neural network technology is largely inspired by artificial neural networks.

[0004] However, in the key issue of obtaining weights, photonic neural networks still rely on artificial neural networks and do not have their own method. The method of obtaining weights by ANN is based on the back propagation algorithm. This method is not only extremely computationally intensive, but also energy-intensive, requiring a large amount of computing resources and energy consumption to achieve its complex operation and processing requirements, resulting in high energy consumption in the image classification process realized by photonic neural networks, which is contrary to the expectation of low power consumption of ONN.

[0005] Therefore, the prior art still needs to be improved and developed. SUMMARY

[0006] The main purpose of the present application is to provide an image classification method and system based on DCPI resin, a terminal and a computer readable storage medium, which aims to solve the problem that photonic neural networks lack a technology to obtain weights with low power consumption in the prior art, resulting in high energy consumption in the image classification process realized by photonic neural networks.

[0007] To achieve the above purpose, the present application provides an image classification method based on DCPI resin, which comprises the following steps:

[0008] A photonic neural network based on DCPI resin is constructed, which comprises an input layer, an optical weight self-learning layer and an output layer.

[0009] training the photonic neural network to obtain a target network;

[0010] collecting a target image, inputting the target image to an input layer of the target network for preprocessing to obtain a preprocessed image, inputting the preprocessed image to an optical weight self-learning layer of the target network for inference to obtain an inference result, inputting the inference result to an output layer of the target network for classification, and outputting a classification result of the target image.

[0011] Optionally, the DCPI resin-based image classification method, wherein the optical weight self-learning layer comprises a UV projection module, a first microlens array module, a DCPI resin module, a second microlens array module, and a laser irradiation module.

[0012] The DCPI resin module is located between the first microlens array module and the second microlens array module.

[0013] Optionally, the DCPI resin-based image classification method, wherein the DCPI resin module is obtained by simulating a biological learning process.

[0014] Optionally, the DCPI resin-based image classification method, wherein the training of the photonic neural network to obtain a target network specifically comprises:

[0015] obtaining a sample image, preprocessing the sample image to obtain a preprocessed sample image;

[0016] inputting the preprocessed sample image to a UV projection module of the optical weight self-learning layer, irradiating the preprocessed sample image onto the DCPI resin module after passing through the first microlens array module by first-wavelength laser emitted by the UV projection module to obtain a first stimulation signal;

[0017] irradiating second-wavelength laser emitted by the laser irradiation module onto the DCPI resin module after passing through the second microlens array module to obtain a second stimulation signal;

[0018] associating the first stimulation signal and the second stimulation signal by the DCPI resin module to automatically generate weights carrying information on the image, and obtaining the target network according to the weights.

[0019] Optionally, the DCPI resin-based image classification method, wherein the inputting of the target image to the input layer of the target network for preprocessing to obtain a preprocessed image specifically comprises:

[0020] The target image is input to the input layer of the target network for noise reduction, filtering and normalization processing to obtain first intermediate data;

[0021] The first intermediate data is compressed and irrelevant information is removed to reduce the number of weights to be learned by the optical weight self-learning layer, to obtain second intermediate data;

[0022] The size of the second intermediate data is scaled according to the size of the optical weight self-learning layer to obtain third intermediate data, and initial feature extraction is performed on the third intermediate data to obtain the preprocessed image.

[0023] Optionally, the DCPI resin-based image classification method, wherein the preprocessed image is input to the optical weight self-learning layer of the target network for inference to obtain an inference result, specifically comprising:

[0024] The preprocessed image is input to the optical weight self-learning layer with weights, and the UV projection module emits a ray to perform super-shooting on the preprocessed image to obtain a green fluorescent signal;

[0025] The fluorescent signal is converted into an electrical signal, and the inference result is obtained according to the electrical signal.

[0026] Optionally, the DCPI resin-based image classification method, wherein the first wavelength laser emitted by the UV projection module is a purple laser with a wavelength of 365nm, and the second wavelength laser emitted by the laser irradiation module is a green laser with a wavelength of 532nm.

[0027] In addition, in order to achieve the above-mentioned purpose, the present application also provides a DCPI resin-based image classification method system, wherein the DCPI resin-based image classification method system comprises:

[0028] A model construction module is configured to construct a DCPI resin-based photonic neural network, wherein the photonic neural network comprises an input layer, an optical weight self-learning layer and an output layer;

[0029] A model training module is configured to train the photonic neural network to obtain a target network;

[0030] An optical computing module is configured to collect a target image, input the target image to the input layer of the target network for preprocessing to obtain a preprocessed image, input the preprocessed image to the optical weight self-learning layer of the target network for inference to obtain an inference result, input the inference result to the output layer of the target network for classification, and output the classification result of the target image.

[0031] In addition, to achieve the above object, the present application also provides a terminal, wherein the terminal comprises a memory, a processor, and a DCPI resin-based image classification method program stored in the memory and executable on the processor, and the DCPI resin-based image classification method program implements the steps of the DCPI resin-based image classification method when executed by the processor.

[0032] In addition, to achieve the above object, the present application also provides a computer readable storage medium, wherein the computer readable storage medium stores a DCPI resin-based image classification method program, and the DCPI resin-based image classification method program implements the steps of the DCPI resin-based image classification method when executed by a processor.

[0033] In the present application, a DCPI resin-based photonic neural network is constructed, which comprises an input layer, an optical weight self-learning layer, and an output layer; the photonic neural network is trained to obtain a target network; a target image is collected and input to the input layer of the target network for preprocessing to obtain a preprocessed image, and the preprocessed image is input to the optical weight self-learning layer of the target network for inference to obtain an inference result, which is input to the output layer of the target network for classification to output a classification result of the target image. In the present application, the DCPI resin-based photonic neural network is an independent photonic neural network that does not depend on artificial neural networks, and the generation of weights does not require a large amount of computing resources and energy consumption, which is an important step of artificial intelligence. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 is a flowchart of a preferred embodiment of the DCPI resin-based image classification method of the present application;

[0035] Figure 2 is a schematic diagram of the optical weight generation and inference process in the DCPI resin-based image classification method of the present application;

[0036] Figure 3 is another schematic diagram of the optical weight generation and inference process in the DCPI resin-based image classification method of the present application;

[0037] Figure 4 is a schematic diagram of irradiating the DCPI resin with light of different wavelengths in the DCPI resin-based image classification method of the present application;

[0038] Figure 5 is a structure diagram of a preferred embodiment of the DCPI resin-based image classification method system of the present application;

[0039] Figure 6The schematic diagram of the operation environment of the preferred embodiment of the terminal of the present application. DETAILED DESCRIPTION

[0040] The present application provides a DCPI resin-based image classification method, system and terminal. To make the purpose, technical solutions and effects of the present application clearer and more explicit, the present application is further described in detail below with reference to the drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0041] Those skilled in the art can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as generally understood by those skilled in the art to which the present application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood as having meanings consistent with those in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as such.

[0042] In addition, if the present application embodiments involve descriptions such as "first", "second", etc., the descriptions of "first", "second", etc. are only for descriptive purposes and should not be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features with "first" and "second" can be explicitly or implicitly included at least one of the features. In addition, the technical solutions of various embodiments can be combined with each other, but it must be based on the realization of ordinary skilled in the art, when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, nor within the scope of protection claimed by the present application.

[0043] The DCPI resin-based image classification method according to the preferred embodiment of the present application, as shown in the figure, includes the following steps: Figure 1

[0044] Step S10, a DCPI resin-based photon neural network is constructed, which includes an input layer, an optical weight self-learning layer and an output layer.

[0045] It can be understood that a complete photon neural network should have three modules of input, processing and output. In the present embodiment, the input layer and the output layer are already built, and the focus of the present application is the processing module, i.e. the optical weight self-learning layer.

[0046] As shown in the figure, the DCPI resin-based photon neural network includes an input layer, an optical weight self-learning layer and an output layer. Figure 2 ​As shown, the optical weight self-learning layer is used for weight self-learning and inference, and the optical weight self-learning layer comprises a UV projection module, a first microlens array module, a DCPI resin module, a second microlens array module and a laser irradiation module; wherein the DCPI resin module is located between the first microlens array module and the second microlens array module.

[0047] Wherein, the UV(Ultraviolet) projection module and the laser irradiation module are used to emit rays of different wavelengths, the first microlens array module and the second microlens array module are used to receive the emitted rays and focus and image them, and the DCPI resin(Dual Color Photo initiator) module is used to receive light from two different directions and associate the stimulation of the double light with the fluorescence spectrum change of the resin, thereby realizing automatic generation of weights.

[0048] Further, the DCPI resin module is obtained by simulating a biological learning process; wherein the simulating a biological learning process is using a double light fluorescence effect to analog Pavlov's dog experiment, realizing Hebb's rule through optical experiments, associating the stimulation of the double light with the fluorescence spectrum change of the resin to realize automatic generation of weights.

[0049] It can be understood that for the acquisition of weights, the present application proposes a more aggressive method, which directly simulates the learning process of biology, such as Pavlov's dog experiment. This is a classic study of conditioned reflexes, which shows how animals can establish conditioned reflexes through learning. Knowing that dogs will instinctively secrete saliva when they see food, Pavlov eventually made the dog secrete saliva when it heard the bell even without food by ringing the bell before each feeding, which proved that animals can form new behavior patterns through repeated stimulus-response processes. This process involves the brain associating two originally independent stimuli in the memory system, so that the response originally triggered by a specific stimulus can be triggered by another new and independent stimulus. These stimuli can enter the brain through different sensory channels, such as vision and hearing, and after complex neural processing, they are linked in the brain, and finally the brain can flexibly integrate information from different sources into a unified memory representation. The memory formed by the brain associating the two is the weight in the physical sense of the present application.

[0050] Step S20, training the photonic neural network to obtain a target network.

[0051] Specifically, a sample image is acquired, and the sample image is preprocessed to obtain a preprocessed sample image.

[0052] In the embodiment, the sample image is input to the input layer of the target network for noise reduction, filtering and normalization processing, and the processed data is compressed and irrelevant information is removed to reduce the number of weights to be learned by the optical weight self-learning layer. The size of the data after removing irrelevant information is scaled according to the size of the optical weight self-learning layer, and the scaled data is subjected to initial feature extraction to obtain a preprocessed sample image.

[0053] As shown in Figure 2 The preprocessed sample image is input to the UV projection module of the optical weight self-learning layer. The first wavelength laser emitted by the UV projection module irradiates the preprocessed sample image after passing through the first microlens array module onto the DCPI resin module to obtain a first stimulation signal. The second wavelength laser emitted by the laser irradiation module irradiates the DCPI resin module after passing through the second microlens array module to obtain a second stimulation signal.

[0054] In the embodiment, the key of the model lies in two independent stimulation signals (the first stimulation signal and the second stimulation signal), and a hardware (DCPI resin) having the ability to simultaneously associate two independent input signals. Specifically, the first wavelength laser emitted by the UV projection module irradiates the preprocessed sample image after passing through the first microlens array module onto the DCPI resin module, and the second wavelength laser emitted by the laser irradiation module irradiates the DCPI resin module after passing through the second microlens array module to obtain a second stimulation signal.

[0055] The first wavelength laser emitted by the UV projection module is a purple laser with a wavelength of 365 nm, and the second wavelength laser emitted by the laser irradiation module is a green laser with a wavelength of 532 nm.

[0056] Further, the first stimulation signal and the second stimulation signal are associated by the DCPI resin module to automatically generate weights carrying information on the image, and the target network is obtained according to the weights.

[0057] Understandably, the two-light fluorescence effect based on DCPI resin exhibits excellent correlation characteristics. Specifically, when DCPI resin is irradiated with a 532nm wavelength (green light) laser, the elicited response appears to the human eye as green light, which can be represented as an unconditioned stimulus (a dog salivates upon seeing food). When the DCPI resin is irradiated with a 365nm wavelength light, the elicited response appears to the human eye as red light, which can be represented as a neutral stimulus (a dog does not salivate upon hearing a bell). However, when the DCPI resin is simultaneously irradiated with both 365nm and 532nm wavelengths, and then irradiated again with 365nm wavelength light, the elicited response appears to the human eye as green light. At this point, the stimuli of the two lights (unconditioned and neutral) are correlated (a dog salivates upon hearing a bell). Therefore, DCPI resin can be considered a correlation element. Analogous to Pavlov's experiment with the two-light fluorescence effect, Heb's law was successfully implemented through optical experiments, correlating the two-light stimulus to changes in the resin's fluorescence spectrum, thereby achieving automatic weight generation and enabling the completion of a typical artificial intelligence task—pattern recognition.

[0058] Furthermore, such as Figure 3 As shown, in another embodiment, the sample to be learned consists of the letters NVZ. These three letters are projected onto the DCPI using a 365nm (purple) projector. When projecting the letter V, only the upper region receives 532nm (green) light, while the middle and lower regions do not. Because only the upper region of the resin receives both 532nm and 365nm light, a change in the fluorescence spectrum occurs. Therefore, when only 365nm light is applied, only the upper region shows a red V. Continuing with the 365nm projector, the letter N is projected. Only the middle region receives 532nm light, while the upper and lower regions do not. Again, because only the middle region of the resin receives both 532nm and 365nm light, a change in the fluorescence spectrum occurs. Therefore, when only 365nm light is applied, only the middle region shows a red N. Continuing to project the letter Z using a 365nm projector, only the lower area receives 532nm light, while the upper and middle areas do not. Because only the lower resin area receives both 532nm and 365nm light, a change in fluorescence spectrum occurs. Therefore, when only 365nm light is applied, only the lower area shows the red letter N. After these steps, the weights of the three letters are engraved on the DCPI.

[0059] It can be understood that the traditional computer digital weight is 0 and 1. In the present application, the weight is distinguished by the excited light, for example: exciting green light is 1, exciting red light is 0, the DCPI is excited to red light 0 when only 365nm light is irradiated, and when 365+532 simultaneously irradiate the DCPI for a period of time, then use 365 alone to irradiate, at this time the DCPI is excited to green light 1, use 365 light to project the letter, such as projecting the letter V, at this time the DCPI will have a red pattern of a V letter appear, then use a 532 light to cover the entire letter area, 365+532 irradiate for a period of time, then use 365 light alone to irradiate, at this time a green pattern of a V letter appears. Thus, the light excited by the DCPI is changed from red light 0 to green light 1, and the weight corresponding to the letter can be engraved on the DCPI by this method.

[0060] At this time, continue to project N, V, Z, the three letters, respectively, using the 365nm projector, according to the DCPI resin after the change of the fluorescence spectrum irradiated by the 365nm light alone, it can be known that the green fluorescence excited is the largest when the N letter is projected. When the V letter is projected, the green fluorescence excited corresponding to the weight of the V letter is the largest. When the Z letter is projected, the green fluorescence excited corresponding to the weight of the Z letter is the largest. Therefore, by analogy with the Pavlov and dog experiment using double light fluorescence effect, the Hebb rule is successfully realized by optical experiment, the stimulation of double light is associated with the change of fluorescence spectrum of the resin, thereby realizing the automatic generation of weight, and the typical artificial intelligence task of pattern recognition can be completed.

[0061] Step S30, collecting a target image, inputting the target image to an input layer of the target network for preprocessing to obtain a preprocessed image, inputting the preprocessed image to an optical weight self-learning layer of the target network for inference to obtain an inference result, inputting the inference result to an output layer of the target network for classification, and outputting a classification result of the target image.

[0062] The inputting of the target image to the input layer of the target network for preprocessing to obtain a preprocessed image specifically includes:

[0063] The target image is input to the input layer of the target network for noise reduction, filtering and normalization processing to obtain first intermediate data;

[0064] The first intermediate data is compressed and irrelevant information is removed to reduce the number of weights to be learned by the optical weight self-learning layer to obtain second intermediate data;

[0065] The size of the second intermediate data is scaled according to the size of the optical weight self-learning layer to obtain third intermediate data, and initial feature extraction is performed on the third intermediate data to obtain the preprocessed image.

[0066] In the embodiment, after the target image is acquired, the target image is first input to the input layer of the target network for preliminary data preprocessing: image data fast preprocessing (such as noise reduction, filtering, normalization), compression of irrelevant information, and reduction of the number of weights to be learned by the subsequent optical module. The image is scaled to adapt to the size of the optical weight self-learning module, and initial features are extracted.

[0067] Further, the preprocessed image is input to the optical weight self-learning layer of the target network for inference to obtain an inference result, specifically including:

[0068] The preprocessed image is input to the optical weight self-learning layer with weights, and the UV projection module emits a ray to perform super-shooting on the preprocessed image to obtain a green fluorescent signal;

[0069] The fluorescent signal is converted into an electrical signal, and the inference result is obtained according to the electrical signal.

[0070] It can be understood that after the photonic neural network is trained, the optical module of the target network at this time has carried information weights, and then UV light is used for irradiation to complete the calculation. This process embodies the speed superiority of optical calculation. When the UV light irradiates the optical module with weights, the picture information appears in the form of green fluorescence, and the green fluorescent signal is collected and converted into an electrical signal to be transmitted to the edge computing platform.

[0071] As shown in Figure 4 When only 365nm light irradiates the DCPI resin, red fluorescence is excited. When only 532nm light irradiates the DCPI resin, green fluorescence is excited. When 365nm and 532nm light irradiate the DCPI resin at the same time, the DCPI resin changes the fluorescence spectrum, and then the DCPI resin after the fluorescence spectrum changes is irradiated with 365nm light alone, and green fluorescence is excited.

[0072] Further, the edge computing platform can perform further logical operations such as classification, prediction, etc. according to the feature data generated by the optical module, so as to improve the reliability of the calculation. In addition to the speed advantage of optical calculation, the combination of edge computing also combines the confidentiality of edge computing. For example, in the product line, the product photo is highly confidential, the characteristic information of the product photo is stored in the DCPI resin of the application, and the DCPI resin with the characteristics is distributed to each production line, and the edge computing platform further processes the subsequent operation according to the feature data (excited fluorescence intensity) generated by the optical module (DCPI), and all operation results are stored locally without uploading to the cloud; the combination of optical weight self-learning and edge computing in real time can greatly improve the task processing efficiency, and the local processing and encryption technology can enhance the data security. Whether in industry, medical treatment or other fields, this architecture can meet the needs of high efficiency and privacy protection at the same time.

[0073] It can be seen that the application constructs a photonic neural network based on DCPI resin, the photonic neural network includes an input layer, an optical weight self-learning layer and an output layer; the photonic neural network is trained to obtain a target network; a target image is collected, the target image is input to the input layer of the target network for preprocessing to obtain a preprocessed image, the preprocessed image is input to the optical weight self-learning layer of the target network for inference to obtain an inference result, and the inference result is input to the output layer of the target network for classification to output a classification result of the target image. The photonic neural network based on DCPI resin in the application is an independent photonic neural network that does not depend on an artificial neural network, and the generation of weights does not require a large amount of computing resources and energy consumption, which is an important step of artificial intelligence.

[0074] Further, as shown in the above-mentioned DCPI resin-based image classification method, the application also correspondingly provides a DCPI resin-based image classification method system, wherein the DCPI resin-based image classification method system comprises: Figure 5 A model construction module 51 is configured to construct a photonic neural network based on DCPI resin, and the photonic neural network includes an input layer, an optical weight self-learning layer and an output layer;

[0075] A model training module 52 is configured to train the photonic neural network to obtain a target network;

[0076]

[0077] ​The optical computing module 53 is configured to collect a target image, input the target image into an input layer of the target network for preprocessing to obtain a preprocessed image, input the preprocessed image into an optical weight self-learning layer of the target network for inference to obtain an inference result, input the inference result into an output layer of the target network for classification, and output a classification result of the target image.

[0078] Further, as shown in Figure 6 Based on the above-mentioned DCPI resin-based image classification method and system, the application also provides a terminal, which comprises a processor 10, a memory 20 and a display 30. Figure 6 Only some components of the terminal are shown, but it should be understood that all the shown components are not required, and more or fewer components can be alternatively implemented.

[0079] The memory 20 can be an internal storage unit of the terminal in some embodiments, such as a hard disk or a memory of the terminal. The memory 20 can also be an external storage device of the terminal in other embodiments, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal. Further, the memory 20 can include both the internal storage unit and the external storage device of the terminal. The memory 20 is used to store application software and various data installed on the terminal, such as program codes of the terminal, etc. The memory 20 can also be used to temporarily store data that has been output or will be output. In an embodiment, the memory 20 stores a DCPI resin-based image classification method program 40, which can be executed by the processor 10 to implement the DCPI resin-based image classification method in the application.

[0080] The processor 10 can be a central processing unit (CPU), a microprocessor or other data processing chip in some embodiments, which is used to run program codes or process data stored in the memory 20, such as to execute the DCPI resin-based image classification method, etc.

[0081] The display 30 can be an LED display, a liquid crystal display, a touch liquid crystal display, an OLED (Organic Light-Emitting Diode) touch, etc. in some embodiments. The display 30 is used to display information of the terminal and to display a visualized user interface. The components of the terminal communicate with each other through a system bus.

[0082] In an embodiment, the following steps are implemented when the processor 10 executes the DCPI resin-based image classification method program 40 in the memory 20:

[0083] A DCPI resin-based photonic neural network is constructed, which includes an input layer, an optical weight self-learning layer, and an output layer;

[0084] The photonic neural network is trained to obtain a target network;

[0085] A target image is collected, preprocessed in the input layer of the target network to obtain a preprocessed image, and input into the optical weight self-learning layer of the target network for inference to obtain an inference result, which is input into the output layer of the target network for classification to output the classification result of the target image.

[0086] The optical weight self-learning layer includes a UV projection module, a first microlens array module, a DCPI resin module, a second microlens array module, and a laser irradiation module.

[0087] The DCPI resin module is located between the first microlens array module and the second microlens array module.

[0088] The DCPI resin module is obtained by simulating a biological learning process.

[0089] The training of the photonic neural network to obtain a target network specifically includes:

[0090] A sample image is obtained and preprocessed to obtain a preprocessed sample image;

[0091] The preprocessed sample image is input into the UV projection module of the optical weight self-learning layer, and the first wavelength laser emitted by the UV projection module is used to irradiate the preprocessed sample image onto the DCPI resin module after passing through the first microlens array module to obtain a first stimulation signal.

[0092] The second wavelength laser emitted by the laser irradiation module is used to irradiate the DCPI resin module after passing through the second microlens array module to obtain a second stimulation signal.

[0093] The first stimulation signal and the second stimulation signal are associated by the DCPI resin module to automatically generate weights carrying information on the image, and the target network is obtained according to the weights.

[0094] The target image is input to the input layer of the target network for preprocessing to obtain a preprocessed image, and the preprocessing specifically includes:

[0095] The target image is input to the input layer of the target network for noise reduction, filtering and normalization processing to obtain first intermediate data.

[0096] The first intermediate data is compressed and irrelevant information is removed to reduce the number of weights to be learned by the optical weight self-learning layer, and second intermediate data is obtained.

[0097] The size of the second intermediate data is scaled according to the size of the optical weight self-learning layer to obtain third intermediate data, and initial feature extraction is performed on the third intermediate data to obtain the preprocessed image.

[0098] The preprocessed image is input to the optical weight self-learning layer of the target network for inference to obtain an inference result, and the inference specifically includes:

[0099] The preprocessed image is input to the optical weight self-learning layer with weights, and the UV projection module emits a ray to perform super-shooting on the preprocessed image to obtain a green fluorescent signal.

[0100] The fluorescent signal is converted into an electrical signal, and the inference result is obtained according to the electrical signal.

[0101] The first wavelength laser emitted by the UV projection module is a purple laser with a wavelength of 365nm, and the second wavelength laser emitted by the laser irradiation module is a green laser with a wavelength of 532nm.

[0102] The application also provides a computer readable storage medium, wherein the computer readable storage medium stores a DCPI resin-based image classification method program, and the DCPI resin-based image classification method program is executed by a processor to realize the steps of the DCPI resin-based image classification method.

[0103] In summary, the application provides an image classification method, system and terminal based on DCPI resin, the method comprising: constructing a photonic neural network based on DCPI resin, the photonic neural network comprising an input layer, an optical weight self-learning layer and an output layer; training the photonic neural network to obtain a target network; collecting a target image, inputting the target image into the input layer of the target network for preprocessing to obtain a preprocessed image, inputting the preprocessed image into the optical weight self-learning layer of the target network for inference to obtain an inference result, inputting the inference result into the output layer of the target network for classification, and outputting a classification result of the target image. The photonic neural network based on DCPI resin in the application is an independent photonic neural network that does not depend on artificial neural networks, and the generation of weights does not require a large amount of computing resources and energy consumption, which is an important step of artificial intelligence.

[0104] It should be noted that in this document, the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusions, so that a process, method, article or terminal including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article or terminal. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of another identical element in the process, method, article or terminal comprising the element.

[0105] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the computer program can include the processes of the above-mentioned embodiments of the method. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0106] It should be understood that the application is not limited to the above examples, and those skilled in the art can make improvements or changes according to the above description, and all these improvements and changes shall fall within the protection scope of the appended claims of the present application.

Claims

1. A method of image classification based on DCPI resin, characterized by, The image classification method based on the DCPI resin comprises the following steps: A photonic neural network based on the DCPI resin is constructed, which comprises an input layer, an optical weight self-learning layer and an output layer; The photonic neural network is trained to obtain a target network; A target image is collected, and the target image is input into the input layer of the target network for preprocessing to obtain a preprocessed image, the preprocessed image is input into the optical weight self-learning layer of the target network for inference to obtain an inference result, the inference result is input into the output layer of the target network for classification, and a classification result of the target image is output; The optical weight self-learning layer comprises a UV projection module, a first microlens array module, a DCPI resin module, a second microlens array module and a laser irradiation module; The DCPI resin module is located between the first microlens array module and the second microlens array module; The photonic neural network is trained to obtain a target network, specifically comprising the following steps: A sample image is obtained, and the sample image is preprocessed to obtain a preprocessed sample image; The preprocessed sample image is input into the UV projection module of the optical weight self-learning layer, and a first wavelength laser emitted by the UV projection module is used to irradiate the preprocessed sample image onto the DCPI resin module through the first microlens array module to obtain a first stimulation signal; A second wavelength laser emitted by the laser irradiation module is used to irradiate the DCPI resin module through the second microlens array module to obtain a second stimulation signal; The first stimulation signal and the second stimulation signal are associated by the DCPI resin module to automatically generate weights carrying information on the image, and the target network is obtained according to the weights.

2. The DCPI resin-based image classification method according to claim 1, characterized in that, The DCPI resin module is obtained by simulating a biological learning process.

3. The DCPI resin-based image classification method according to claim 1, characterized in that, The target image is input into the input layer of the target network for preprocessing to obtain a preprocessed image, specifically comprising the following steps: The target image is input into the input layer of the target network for noise reduction, filtering and normalization processing to obtain first intermediate data; The first intermediate data is compressed and irrelevant information is removed to reduce the number of weights to be learned by the optical weight self-learning layer to obtain second intermediate data; The size of the second intermediate data is scaled according to the size of the optical weight self-learning layer to obtain third intermediate data, and initial feature extraction is performed on the third intermediate data to obtain the preprocessed image.

4. The DCPI resin-based image classification method according to claim 1, characterized by, The preprocessed image is input into the optical weight self-learning layer of the target network for inference to obtain an inference result, specifically comprising the following steps: The preprocessed image is input into the optical weight self-learning layer with weights, and the preprocessed image is superimposed by rays emitted by the UV projection module to obtain a green fluorescent signal; The fluorescent signal is converted into an electrical signal, and the inference result is obtained according to the electrical signal.

5. The DCPI resin-based image classification method according to claim 1, characterized in that, The first wavelength laser emitted by the UV projection module is a purple laser with a wavelength of 365nm, and the second wavelength laser emitted by the laser irradiation module is a green laser with a wavelength of 532nm.

6. A DCPI resin based image classification method system characterized by, The DCPI resin-based image classification method system is applied to the DCPI resin-based image classification method of any one of claims 1-5, and the DCPI resin-based image classification method system comprises: a model construction module, configured to construct a DCPI resin-based photon neural network, the photon neural network comprising an input layer, an optical weight self-learning layer, and an output layer; a model training module, configured to train the photon neural network to obtain a target network; an optical calculation module, configured to collect a target image, input the target image into the input layer of the target network for preprocessing to obtain a preprocessed image, input the preprocessed image into the optical weight self-learning layer of the target network for inference to obtain an inference result, input the inference result into the output layer of the target network for classification, and output a classification result of the target image.

7. A terminal, characterized by comprising: The terminal comprises a memory, a processor, and a DCPI resin-based image classification method program stored on the memory and executable on the processor, and the DCPI resin-based image classification method program, when executed by the processor, implements the steps of the DCPI resin-based image classification method of any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a DCPI resin-based image classification method program, and the DCPI resin-based image classification method program, when executed by the processor, implements the steps of the DCPI resin-based image classification method of any one of claims 1-5.

Images