A frosting area recognition method, device and equipment based on pulse neural network

Abstract

The present invention relates to the field of refrigeration equipment maintenance technology, and in particular to a frosting area recognition method, device and equipment based on a pulse neural network. The method comprises: performing a frame difference operation on an image of an air cooler to obtain a frame difference image; inputting the grayscale value of each pixel point of the frame difference image into a pulse neural network, deepening the difference of neuron pulse signals through the dynamic connection strength between a coding layer and an excitation layer, and obtaining the pulse firing rate of each neuron; forming a frequency interval with the maximum and minimum frequency values ​​in each pulse firing rate, eliminating the high frequency band and the low frequency band of the frequency interval at a preset ratio, and obtaining the excited pixel points corresponding to the neurons whose pulse firing rates are in the remaining frequency bands; obtaining the frosting area according to the coordinates of the excited pixel points, and accurately and efficiently obtaining the frosting area of ​​the air cooler.

Description

Technical Field

[0001] The present invention relates to the technical field of refrigeration equipment maintenance, and in particular to a frosting area recognition method, device and equipment based on a pulse neural network. Background Art

[0002] After the air cooler has been running for a period of time, its evaporator will often be frosted, and the frosted surface will affect the cooling efficiency of the air cooler, so the air cooler needs to be defrosted.

[0003] When obtaining the frosting condition of the air cooler through the image of the air cooler and performing defrosting, the area outside the frosted area of ​​the air cooler will affect the recognition effect of the frosting condition of the air cooler image, resulting in a large error; therefore, it is necessary to divide the area of ​​the air cooler that the camera is aimed at, and manually set the calculation area in advance according to the camera position.

[0004] However, this method is limited by the position of the camera. Once the camera sends an offset, the frosting area needs to be readjusted, which increases the debugging workload. On the other hand, the manual setting of the frosting area is subjective, because the frosting of the air cooler evaporator is a relatively random process. If the calculation area is not set reasonably, the correlation between the calculation result and the actual frosting degree will be greatly reduced. Summary of the invention

[0005] The present application provides a frosting area recognition method, device and equipment based on a pulse neural network, which solves the problem of inaccurate and low-efficiency setting of frosting areas in the prior art.

[0006] The first aspect of the present application provides a frosting area recognition method based on a pulse neural network, comprising:

[0007] Performing a frame difference operation on the air cooler image to obtain a frame difference image;

[0008] The frame difference image is input into a spiking neural network, which includes: a coding layer and an excitation layer; the coding layer converts the frame difference image into a pulse sequence, and sends a pulse signal to the excitation layer according to the pulse sequence, and the network calculates the pulse firing rate of each neuron in the excitation layer; the connection strength between the coding layer and the excitation layer changes dynamically according to the pulse firing situation at the previous moment, increasing the connection strength of the active neuron synapse and reducing the connection strength of the inactive neuron synapse;

[0009] The maximum and minimum frequencies in each pulse firing rate form a frequency interval, the high frequency band and the low frequency band of the frequency interval are eliminated at a preset ratio, and the excited pixel points corresponding to the neurons whose pulse firing rates are in the remaining frequency bands are obtained;

[0010] The frosting area is obtained according to the coordinates of the excited pixel points.

[0011] Optionally, the connection strength between the coding layer and the excitation layer changes dynamically according to the pulse emission at the previous moment:

[0012] Each neuron in the coding layer is connected to each neuron in the excitation layer one by one, and the connection strength between the coding layer and the excitation layer is dynamically updated using the STDP rule;

[0013] The STDP rules are specifically:

[0014]

[0015] Among them, b and a represent the weight update rate and the weight dependence on the previous moment respectively, △y is the connection strength increment between synapses, which is continuously updated with the release of synaptic neuron pulses, y max is the maximum upper limit of synaptic connection strength, and y is the connection strength between synapses. p represents the presynaptic trace, u1 is the presynaptic trace when the neuron emits a pulse, β up is the time constant.

[0016] Optionally, the spiking neural network further includes: an inhibition layer;

[0017] Each neuron in the inhibition layer is connected to each neuron in the excitation layer, and a preset fixed connection strength is used to send an inhibition signal to the neurons in the excitation layer that have not sent a pulse signal.

[0018] Optionally, the step of obtaining excited pixel points corresponding to neurons whose pulse firing rates are in the remaining frequency bands is specifically as follows:

[0019] The remaining frequency segments are evenly divided into a preset number of intervals. According to the number of neurons in each interval, the first interval in which the number of neurons decreases with increasing frequency and the preset number of intervals after this interval are selected; the pixel points corresponding to the neurons in the selected interval are identified to obtain the excited pixel points.

[0020] Optionally, before performing the frame difference operation on the air cooler image, the method further includes:

[0021] The image obtained by the air cooler camera is grayed out and scaled to the size corresponding to the spiking neural network so that the number of image pixels corresponds to the number of neurons in the spiking neural network layer.

[0022] Optionally, performing a frame difference operation on the air cooler image to obtain a frame difference image is specifically:

[0023] Subtract the grayscale value of the corresponding pixel point of the current air cooler image from the previous air cooler image in the time sequence, substitute each grayscale difference value into the grayscale difference processing model, substitute the processed grayscale difference value into the corresponding pixel point position, and obtain a frame difference image;

[0024] The grayscale difference processing model is specifically:

[0025]

[0026] Among them, S t (x, y) is the grayscale difference of the pixel with coordinates (x, y) after processing, F t (x, y) is the grayscale difference of the pixel with coordinates (x, y) before processing, minF t Grayscale difference minimum value, maxF t Maximum grayscale difference.

[0027] The second aspect of the present application provides a frosting area recognition device based on a pulse neural network, comprising:

[0028] A frame difference module, used for performing a frame difference operation on the cooling fan image to obtain a frame difference image;

[0029] A neural network module inputs the frame difference image into a pulse neural network, wherein the pulse neural network includes: a coding layer and an excitation layer; the coding layer converts the frame difference image into a pulse sequence, and sends a pulse signal to the excitation layer according to the pulse sequence, and the network calculates the pulse firing rate of each neuron in the excitation layer; the connection strength between the coding layer and the excitation layer changes dynamically according to the pulse firing situation at the previous moment, increasing the connection strength of the active neuron synapse and reducing the connection strength of the inactive neuron synapse;

[0030] A screening module is used to form a frequency interval with the maximum and minimum frequencies in each pulse firing rate, remove the high frequency band and the low frequency band of the frequency interval at a preset ratio, and obtain the excited pixel points corresponding to the neurons whose pulse firing rates are in the remaining frequency bands;

[0031] The frosting area building module is used to obtain the frosting area according to the coordinates of the exciting pixel points.

[0032] Optionally, in the neural network module, the connection strength between the coding layer and the excitation layer changes dynamically according to the pulse emission at the previous moment:

[0033] Each neuron in the coding layer is connected to each neuron in the excitation layer one by one, and the connection strength between the coding layer and the excitation layer is dynamically updated using the STDP rule;

[0034] The STDP rules are specifically:

[0035]

[0036] Among them, b and a represent the weight update rate and the weight dependence on the previous moment respectively, △y is the connection strength increment between synapses, which is continuously updated with the release of synaptic neuron pulses, y max is the maximum upper limit of synaptic connection strength, and y is the connection strength between synapses. p represents the presynaptic trace, u1 is the presynaptic trace when the neuron emits a pulse, β up is the time constant.

[0037] Optionally, in the neural network module, the spiking neural network further includes: an inhibition layer;

[0038] Each neuron in the inhibition layer is connected to each neuron in the excitation layer, and a preset fixed connection strength is used to send an inhibition signal to the neurons in the excitation layer that have not sent a pulse signal.

[0039] The third aspect of the present application provides a frosting area recognition device based on a pulse neural network, the device comprising a processor and a memory:

[0040] The memory is used to store program code and transmit the program code to the processor;

[0041] The processor is used to execute the frosting area recognition method based on pulse neural network as described in any one of the first aspects of the present application according to the instructions in the program code.

[0042] The frosted area recognition method based on pulse neural network provided in the present application obtains the pulse firing rate of each neuron by inputting the frame difference image of the cold air fan image into the pulse neural network, and reflects the gray value change of the corresponding pixel position with the pulse firing rate, deepens the difference of the neuron pulse firing rate with the dynamic connection strength, so that the change of frosting can be accurately identified and quantified, and the frequency maximum and minimum values ​​in each pulse firing rate constitute a frequency interval, and the high frequency band and low frequency band of the frequency interval are eliminated at a preset ratio, and the excited pixel points corresponding to the neurons whose pulse firing rates are in the remaining frequency bands are obtained, so as to accurately distinguish the neurons in the frosted and non-frosted areas, and obtain the frosted area according to the excited pixel points corresponding to the remaining neurons, thereby avoiding the subjectivity of manual area division and improving the efficiency and accuracy of dividing the frosted area. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0044] Figure 1 A schematic diagram of the process of the frosting area recognition method based on the pulse neural network provided in this application;

[0045] Figure 2 A schematic diagram of the pulse emission rate acquisition process of the frosting area identification method based on the pulse neural network provided in this application;

[0046] Figure 3 Schematic diagram of the pulse neural network structure provided for this application;

[0047] Figure 4 A schematic diagram of the exciting pixel point selection process of the frosting area recognition method based on the pulse neural network provided in this application;

[0048] Figure 5 A schematic diagram of the overall process of the frosting area recognition method based on the pulse neural network provided in this application;

[0049] Figure 6 A schematic diagram of the structure of a frosting area identification device based on a pulse neural network provided in this application. DETAILED DESCRIPTION

[0050] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] The present application provides a frosting area recognition method based on a pulse neural network, which solves the problem of low efficiency and poor accuracy in the prior art of manually dividing the frosting area of ​​the air cooler.

[0052] See also Figure 1 , Figure 1 A schematic flow chart of a frosting area recognition method based on a pulse neural network provided in this application.

[0053] A first aspect of the present embodiment provides a frosting area recognition method based on a pulse neural network, comprising:

[0054] S100, performing a frame difference operation on the air cooler image to obtain a frame difference image;

[0055] It should be noted that, in this embodiment, images of the air cooler are captured at regular intervals, and the grayscale values ​​of pixels of images of adjacent frames are subtracted to obtain a frame difference image using the grayscale difference.

[0056] S200, inputting the frame difference image into a spiking neural network, the spiking neural network comprising: a coding layer and an excitation layer; the coding layer converts the frame difference image into a pulse sequence, and sends a pulse signal to the excitation layer according to the pulse sequence, and the network calculates the pulse firing rate of each neuron in the excitation layer; the connection strength between the coding layer and the excitation layer changes dynamically according to the pulse firing situation at the previous moment, increasing the connection strength of the active neuron synapse and reducing the connection strength of the inactive neuron synapse;

[0057] It should be noted that the grayscale value of each pixel in the frame difference image is the difference between the grayscale values ​​of each pixel in adjacent frames, which represents the change in the frost condition of the air cooler position corresponding to the pixel in the captured air cooler image; the greater the change in the frost condition, the greater the grayscale value of the corresponding frame difference image;

[0058] By dynamically changing the connection strength between the encoding layer and the excitation layer, active neurons become more susceptible to pulse signals by increasing the synaptic connection strength, while inactive neurons become less susceptible to pulse signals by reducing the synaptic connection strength, making the final difference in the pulse firing rate of neurons more significant, making it easier to distinguish between frosted areas and non-frosted areas.

[0059] Furthermore, the spiking neural network calculates the pulse firing rate of each neuron in the excitatory layer.

[0060] S300, forming a frequency interval with the maximum and minimum frequencies of each pulse firing rate, removing the high frequency band and the low frequency band of the frequency interval at a preset ratio, and obtaining excited pixel points corresponding to neurons whose pulse firing rates are in the remaining frequency bands;

[0061] It should be noted that, according to the interval formed by each pulse emission rate, the eliminated low-frequency band is the area where the pulse signal is not obvious, which corresponds to the non-frosted area, and the eliminated high-frequency end is the abnormal pulse signal, which is interference noise; the staff can set the ratio of the eliminated low-frequency band and high-frequency end according to the actual use of the air cooler.

[0062] The position of the neuron in the excitation layer corresponds to the position of the pixel in the image, that is, the position of the air cooler; the pixel corresponding to the neuron whose pulse emission rate is in the remaining frequency range is the excited pixel, and frost will appear at the corresponding air cooler position.

[0063] S400, obtaining a frosting area according to the coordinates of the excited pixel points;

[0064] It should be noted that one or more frosting areas are formed according to the coordinates of each excited pixel corresponding to the image.

[0065] In this embodiment, the frame difference image of the air cooler image is input into the pulse neural network, and the dynamic connection strength is used to make the pulse firing rate between neurons obviously different. Then the pulse firing rate of each neuron is screened, and the neurons in the non-frosted area are eliminated in the frequency range. Based on the excited pixel points corresponding to the remaining neurons, the frosted area of ​​the air cooler is accurately and efficiently obtained.

[0066] The above is a detailed description of the first embodiment of a frosting area recognition method based on a pulse neural network provided by the present application. The following is a detailed description of the second embodiment of a frosting area recognition method based on a pulse neural network provided by the present application.

[0067] See also Figure 2 This embodiment provides a frosting area recognition method based on a pulse neural network. In step S100 of the aforementioned embodiment, a frame difference operation is performed on the air cooler image to obtain a frame difference image including:

[0068] S110, gray-scale the image acquired by the air cooler camera, and scale the image to a size corresponding to the pulse neural network, so that the number of image pixels corresponds to the number of neurons in the pulse neural network layer.

[0069] It should be noted that the air cooler image captured by the air cooler camera is grayed and then scaled so that the position and number of pixel points in the image can match the neuron settings of the pulse neural network, which is convenient for the corresponding input pulse sequence in the subsequent steps and the image pixel coordinates corresponding to the recognition neurons.

[0070] Furthermore, pixels and neurons can correspond one to one; multiple pixels can correspond to one neuron, and the information of multiple pixels can be input into the same neuron; one pixel can correspond to multiple neurons, and the information of the same pixel can be input into multiple neurons.

[0071] S120, subtract the grayscale values ​​of the corresponding pixels of the current air cooler image from those of the previous air cooler image in time sequence, substitute each grayscale difference into the grayscale difference processing model, substitute the processed grayscale difference into the corresponding pixel position, and obtain a frame difference image.

[0072] It should be noted that the grayscale difference processing model is specifically:

[0073]

[0074] Among them, S t (x, y) is the grayscale difference of the pixel with coordinates (x, y) after processing, F t (x, y) is the grayscale difference of the pixel with coordinates (x, y) before processing, minF t Grayscale difference minimum value, maxF t Maximum grayscale difference.

[0075] The frame difference image is a grayscale image, and its grayscale value should be between 0 and 255. After the frame difference operation is performed on the air cooler image, negative grayscale differences may appear, so we need to correct them to positive values. Through the grayscale difference processing model, each grayscale difference is corrected according to the specific situation of the parameters obtained by the frame difference operation, and finally a frame difference image is formed.

[0076] In step S200 of the aforementioned embodiment, the frame difference image is input into a spiking neural network, and the spiking neural network includes: a coding layer and an excitation layer; the coding layer converts the frame difference image into a pulse sequence, and sends a pulse signal to the excitation layer according to the pulse sequence, and the network calculates the pulse firing rate of each neuron in the excitation layer; the connection strength between the coding layer and the excitation layer changes dynamically according to the pulse firing situation at the previous moment, increases the connection strength of the active neuron synapses, and reduces the connection strength of the inactive neuron synapses, specifically including steps S210 and S220:

[0077] S210, inputting the grayscale value of each pixel of the frame difference image into the pulse neural network encoding layer, the encoding layer converts the grayscale value of each pixel of the frame difference image into a pulse sequence, and sends a pulse signal to the excitation layer according to the pulse sequence.

[0078] Please note that, see Figure 3 , Figure 3 This is a schematic diagram of the structure of a pulse neural network. The pulse neural network includes: an encoding layer, an excitation layer, and an inhibition layer.

[0079] The grayscale value data of the frame difference image is input into the encoding layer in the pulse neural network. The encoding layer performs Poisson encoding on the grayscale data in the form of frequency, converts the grayscale value data of the frame difference image into a corresponding pulse sequence, and completes the transformation of the input data from numerical value to pulse form.

[0080] The neurons in the excitation layer are connected one-to-one with the neurons in the coding layer, and the excitation layer receives the pulse signal sent by the coding layer.

[0081] Furthermore, the connection strength between the coding layer and the excitation layer is dynamically updated using the STDP rule to adjust the connection relationship between neuronal synapses, so that the weight of active neurons is increased and the weight of inactive neurons is decreased, so that the pulse signal frequency of active and inactive neurons can be further polarized in the excitation layer, which is more conducive to the accurate identification of frosting areas in subsequent steps; the STDP rule is specifically:

[0082]

[0083] Among them, b and a represent the weight update rate and the weight dependence on the previous moment respectively, △y is the connection strength increment between synapses, which is continuously updated with the release of synaptic neuron pulses, y max is the maximum upper limit of synaptic connection strength, and y is the connection strength between synapses. p represents the presynaptic trace, u1 is the presynaptic trace when the neuron emits a pulse, β up is the time constant.

[0084] Furthermore, the neurons in the excitatory layer are connected one-to-one with the neurons in the inhibitory layer, and each neuron in the inhibitory layer is connected to all neurons in the excitatory layer. The inhibitory layer is used to introduce the lateral inhibition mechanism into the network. The lateral inhibition mechanism sends an inhibitory signal to the excitatory layer neurons that have not fired pulses, which reduces the membrane potential of these neurons and makes it more difficult for them to fire pulses. On the one hand, the lateral inhibition mechanism inhibits the inactive neuron population, and on the other hand, it reduces the impact of interference signals.

[0085] Furthermore, a preset fixed connection strength is used between the excitation layer and the inhibition layer to simplify the correction and adjustment of the inhibition degree, and the connection strength can be set according to the actual use of the air cooler.

[0086] S230, the network calculates the pulse firing rate of each neuron in the excitatory layer.

[0087] The pulse cycle after the input pulse sequence is set, and the number of pulse signals emitted by each neuron in the excitation layer is counted after the pulse neural network runs for this cycle time. The emission frequency is calculated based on the number of pulses emitted and the pulse cycle time to obtain the pulse emission rate of each neuron.

[0088] In this embodiment, the air cooler image is scaled and frame-differenced and input into a pulse neural network. The grayscale value is converted into a pulse sequence by the coding layer, and the neuron signal is received and inhibited by the excitation layer and the inhibition layer. Finally, the pulse firing rate of the neuron corresponding to each pixel is obtained, which reflects the frost condition of the corresponding air cooler area. The pulse neural network is used to objectively and accurately screen the neurons corresponding to the pixel points, which not only improves the efficiency but also increases the recognition accuracy.

[0089] The above is a detailed description of the second embodiment of a frosting area recognition method based on a pulse neural network provided by the present application. The following is a detailed description of the third embodiment of a frosting area recognition method based on a pulse neural network provided by the present application.

[0090] See also Figure 4 This embodiment provides a frosting area recognition method based on a pulse neural network. In step S300 of the aforementioned embodiment, the maximum and minimum frequencies in each pulse firing rate form a frequency interval, the high frequency band and the low frequency band of the frequency interval are eliminated at a preset ratio, and the excited pixel points corresponding to the neurons whose pulse firing rates are in the remaining frequency bands are obtained, which specifically includes:

[0091] S310, forming a frequency interval with the maximum and minimum frequencies of each pulse emission rate, and removing a high frequency segment and a low frequency segment of the frequency interval with a preset ratio.

[0092] It is necessary to explain that before the frequency elimination screening is performed, it can also include judging whether the number of frame difference images acquired and input into the pulse neural network reaches a threshold value. When the threshold value is reached, the frequency screening of step S300 is performed. At this time, the number of neuron pulse firing rates obtained can make the frosting area recognition have a better effect. If the threshold value has not been reached, the frame difference image acquisition and input into the pulse neural network are continued.

[0093] Furthermore, the processing of multiple frame difference images obtained by long-term shooting can more completely and accurately identify the frosted area of ​​the air cooler, and the recognition effect is more precise; the number of images is the number of original images input into the network. Corresponding to the collection of air cooler images, the general collection time period can be set to one day, and the interval time is set to 1 minute or 30 seconds, that is, the number of collected images is 60*24=1440 or 120*24=2880, forming 2879 frame difference images for frosted area recognition. The staff can set the number threshold of frame difference images to 2879.

[0094] By counting the pulse firing rate of each neuron, a frequency interval is formed with the maximum pulse firing rate and the minimum pulse firing rate, and the pulse firing rate of each neuron is within this frequency interval; then the high-frequency band and the low-frequency band of the frequency interval are eliminated according to a preset ratio. The preset ratio is 10% in this embodiment, that is, after the frequency interval is divided into ten equal parts, both the lowest frequency and the highest frequency bands are eliminated; the low-frequency band is where the neuron pulse firing rate is low, corresponding to the position where the frosting change is extremely small, and the area corresponding to this position is the non-frosted area; the high-frequency band corresponds to an extremely high pulse firing rate, and the frosted area will not have such a large frame difference grayscale value change, which is regarded as abnormal or noise data in conversion or image acquisition.

[0095] The staff can set the ratio according to the actual use of the air cooler to make the eliminated low-frequency and high-frequency bands more accurate and reasonable.

[0096] S320, divide the remaining frequency segment into a preset number of partitions on average, select the first partition in which the number of neurons decreases as the frequency increases, and a preset number of partitions after the partition according to the number of neurons in each partition; identify the pixel points corresponding to the neurons in the selected partitions to obtain excited pixel points.

[0097] It should be noted that in this embodiment, a frequency histogram can be formed according to the frequency between each partition and the number of neurons in the frequency between the partitions to assist in the selection of the interval; that is, the frequency is used as the width of the histogram and the number of neurons is used as the length of the histogram to form a frequency histogram; then the first partition in which the number of neurons decreases in the histogram is identified, and the preset number of partitions after it is selected, and the pixels corresponding to the neurons in the selected interval are the excited pixels. The preset number of average partitions and the preset number of selected partitions can be set according to the number of neurons. The more neurons there are, the more preset number of average partitions there are, the more partitions there are selected, that is, the more pixels there are in the corresponding partition identification, the more refined the division and selection of the partitions, and the better the final recognition effect will be.

[0098] The exciting pixel points can be selected using a frosting area screening model, and the frosting area screening model is specifically:

[0099]

[0100] Among them, S is the frequency value range of the excited pixel, N is the corresponding frequency value of the excited pixel, and M j With M j+4 is the pulse firing rate covered by the jth to j+4th intervals in the frequency histogram, j is the first interval number where the number of neurons decreases, D i With D i-1 are the numbers of neurons in the i-th and i-1-th frequency partitions among the 10 equally divided frequency partitions.

[0101] Furthermore, when processing multiple frame difference images for a period of time, the above interval screening method is based on the pulse emission rate of multiple frame differences of each neuron. After the actual air cooler works for a period of time, the frosting situation will gradually stabilize over time, that is, the frosted surface of the frosted area will gradually balance between melting and condensation, and the pulse emission rate will also decrease accordingly. However, it can be understood that in the frosting process, the frosted area is approximately close to the center of the area. The greater the frosting change, the smaller the corresponding range. Therefore, the frosting area is reduced and the frosting effect increases, that is, the interval corresponding to the decrease in the number of neurons and the increase in frequency. It has been verified in actual experimental conditions that the pixel points corresponding to the interval before the first interval where the number of neurons decreases are more likely to belong to the non-frosted area.

[0102] For further information, see Figure 5 , Figure 5 A schematic diagram of the overall process of the frosting area recognition method based on the pulse neural network provided in this application. Figure 5 For other steps in the above, please refer to the corresponding process in the above embodiments, which will not be described in detail here.

[0103] In this embodiment, by statistically screening the pulse firing rate of each neuron, neurons corresponding to the non-frosted area and the frequency interval constituting noise are eliminated, and then the remaining frequency intervals are selected to further obtain more accurate excited pixel points.

[0104] The above is a detailed description of the third embodiment of a frosting area recognition method based on a pulse neural network provided by the present application. The following is a detailed description of a frosting area recognition device based on a pulse neural network provided by the second aspect of the present application.

[0105] See also Figure 6 This embodiment provides a frosting area recognition device based on a pulse neural network, comprising:

[0106] A frame difference module 10 is used to perform a frame difference operation on the air cooler image to obtain a frame difference image;

[0107] The neural network module 20 inputs the frame difference image into a pulse neural network, and the pulse neural network includes: a coding layer and an excitation layer; the coding layer converts the frame difference image into a pulse sequence, and sends a pulse signal to the excitation layer according to the pulse sequence, and the network calculates the pulse firing rate of each neuron in the excitation layer; the connection strength between the coding layer and the excitation layer changes dynamically according to the pulse firing situation at the previous moment, increasing the connection strength of the active neuron synapse and reducing the connection strength of the inactive neuron synapse;

[0108] A screening module 30 is used to form a frequency interval with the maximum and minimum frequencies of each pulse firing rate, remove the high frequency band and the low frequency band of the frequency interval at a preset ratio, and obtain excited pixel points corresponding to neurons whose pulse firing rates are in the remaining frequency bands;

[0109] The frosting area building module 40 is used to obtain the frosting area according to the coordinates of the exciting pixel points.

[0110] Furthermore, in the neural network module 20, the connection strength between the coding layer and the excitation layer changes dynamically according to the pulse emission at the previous moment:

[0111] Each neuron in the coding layer is connected to each neuron in the excitation layer one by one, and the connection strength between the coding layer and the excitation layer is dynamically updated using the STDP rule;

[0112] The STDP rules are specifically:

[0113]

[0114] Among them, b and a represent the weight update rate and the weight dependence on the previous moment respectively, △y is the connection strength increment between synapses, which is continuously updated with the release of synaptic neuron pulses, y max is the maximum upper limit of synaptic connection strength, and y is the connection strength between synapses. p represents the presynaptic trace, u1 is the presynaptic trace when the neuron emits a pulse, β up is the time constant.

[0115] Further, in the neural network module 20, the spiking neural network further includes: an inhibition layer;

[0116] Each neuron in the inhibition layer is connected to each neuron in the excitation layer, and a preset fixed connection strength is used to send an inhibition signal to the neurons in the excitation layer that have not sent a pulse signal.

[0117] The third aspect of the present application also provides a frost area identification device based on a pulse neural network, including a processor and a memory: the memory is used to store program code and transmit the program code to the processor; the processor is used to execute the above-mentioned frost area identification method based on a pulse neural network according to the instructions in the program code.

[0118] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and equipment can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0119] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0120] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0121] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0122] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0123] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A frosting area recognition method based on pulse neural network, characterized in that: include: Performing a frame difference operation on the air cooler image to obtain a frame difference image; The frame difference image is input into a spiking neural network, which includes: a coding layer and an excitation layer; the coding layer converts the frame difference image into a pulse sequence, and sends a pulse signal to the excitation layer according to the pulse sequence, and the network calculates the pulse firing rate of each neuron in the excitation layer; the connection strength between the coding layer and the excitation layer changes dynamically according to the pulse firing situation at the previous moment, increasing the connection strength of the active neuron synapse and reducing the connection strength of the inactive neuron synapse; The maximum and minimum frequencies in each pulse firing rate form a frequency interval, the high frequency band and the low frequency band of the frequency interval are eliminated at a preset ratio, and the excited pixel points corresponding to the neurons whose pulse firing rates are in the remaining frequency bands are obtained; Obtaining a frosting area according to the coordinates of the excited pixel points; The frame difference operation is performed on the air cooler image to obtain the frame difference image, specifically: Subtract the grayscale value of the corresponding pixel point of the current air cooler image from the previous air cooler image in the time sequence, substitute each grayscale difference value into the grayscale difference processing model, substitute the processed grayscale difference value into the corresponding pixel point position, and obtain a frame difference image; The grayscale difference processing model is specifically: Among them, S t (x, y) is the grayscale difference of the pixel with coordinates (x, y) after processing, F t (x, y) is the grayscale difference of the pixel with coordinates (x, y) before processing, minF t Grayscale difference minimum value, maxF t Maximum grayscale difference; The method of obtaining excited pixel points corresponding to neurons whose pulse firing rates are in the remaining frequency bands is specifically as follows: the remaining frequency bands are evenly divided into a preset number of partitions, and according to the number of neurons in each partition, the first partition in which the number of neurons decreases as the frequency increases, and a preset number of partitions after the partition are selected; the pixel points corresponding to the neurons in the selected partitions are identified to obtain excited pixel points.

2. The frosting area recognition method based on pulse neural network according to claim 1 is characterized in that: The connection strength between the coding layer and the excitation layer changes dynamically according to the pulse emission at the previous moment: Each neuron in the coding layer is connected to each neuron in the excitation layer one by one, and the connection strength between the coding layer and the excitation layer is dynamically updated using the STDP rule; The STDP rules are specifically: Among them, b and a represent the weight update rate and the weight dependence on the previous moment respectively, △y is the connection strength increment between synapses, which is continuously updated with the release of synaptic neuron pulses, y max is the maximum upper limit of synaptic connection strength, y is the connection strength between synapses; u p represents the presynaptic trace, u1 is the presynaptic trace when the neuron emits a pulse, β up is the time constant.

3. The frosting area recognition method based on pulse neural network according to claim 1 is characterized in that: The spiking neural network further comprises: an inhibition layer; Each neuron in the inhibition layer is connected to each neuron in the excitation layer, and a preset fixed connection strength is used to send an inhibition signal to the neurons in the excitation layer that have not sent a pulse signal.

4. The frosting area recognition method based on pulse neural network according to claim 1 is characterized in that: Before performing the frame difference operation on the cooling fan image, the method further includes: The image obtained by the air cooler camera is grayed out and scaled to the size corresponding to the spiking neural network so that the number of image pixels corresponds to the number of neurons in each layer of the spiking neural network.

5. A frosting area recognition device based on pulse neural network, characterized in that: include: A frame difference module, used for performing a frame difference operation on the cooling fan image to obtain a frame difference image; A neural network module inputs the frame difference image into a pulse neural network, wherein the pulse neural network includes: a coding layer and an excitation layer; the coding layer converts the frame difference image into a pulse sequence, and sends a pulse signal to the excitation layer according to the pulse sequence, and the network calculates the pulse firing rate of each neuron in the excitation layer; the connection strength between the coding layer and the excitation layer changes dynamically according to the pulse firing situation at the previous moment, increasing the connection strength of the active neuron synapse and reducing the connection strength of the inactive neuron synapse; A screening module is used to form a frequency interval with the maximum and minimum frequencies in each pulse firing rate, remove the high frequency band and the low frequency band of the frequency interval at a preset ratio, and obtain the excited pixel points corresponding to the neurons whose pulse firing rates are in the remaining frequency bands; A frosting area building module, used to obtain a frosting area according to the coordinates of the excited pixel points; In the frame difference module, a frame difference operation is performed on the air cooler image to obtain a frame difference image, specifically: Subtract the grayscale value of the corresponding pixel point of the current air cooler image from the previous air cooler image in the time sequence, substitute each grayscale difference value into the grayscale difference processing model, substitute the processed grayscale difference value into the corresponding pixel point position, and obtain a frame difference image; The grayscale difference processing model is specifically: Among them, S t (x, y) is the grayscale difference of the pixel with coordinates (x, y) after processing, F t (x, y) is the grayscale difference of the pixel with coordinates (x, y) before processing, minF t Grayscale difference minimum value, maxF t Grayscale difference maximum value; In the screening module, the excited pixel points corresponding to the neurons whose pulse firing rate is in the remaining frequency band are obtained, specifically: the remaining frequency band is evenly divided into a preset number of partitions, and according to the number of neurons in each partition, the first partition in which the number of neurons decreases with increasing frequency and the preset number of partitions after the partition are selected; the pixel points corresponding to the neurons in the selected partitions are identified to obtain the excited pixel points.

6. The frosting area recognition device based on pulse neural network according to claim 5, characterized in that: In the neural network module, the connection strength between the coding layer and the excitation layer changes dynamically according to the pulse emission at the previous moment: Each neuron in the coding layer is connected to each neuron in the excitation layer one by one, and the connection strength between the coding layer and the excitation layer is dynamically updated using the STDP rule; The STDP rules are specifically: Among them, b and a represent the weight update rate and the weight dependence on the previous moment respectively, △y is the connection strength increment between synapses, which is continuously updated with the release of synaptic neuron pulses, y max is the maximum upper limit of synaptic connection strength, y is the connection strength between synapses; u p represents the presynaptic trace, u1 is the presynaptic trace when the neuron emits a pulse, β up is the time constant.

7. The frosting area recognition device based on pulse neural network according to claim 5, characterized in that: In the neural network module, the spiking neural network further comprises: an inhibition layer; Each neuron in the inhibition layer is connected to each neuron in the excitation layer, and a preset fixed connection strength is used to send an inhibition signal to the neurons in the excitation layer that have not sent a pulse signal.

8. A frosting area recognition device based on a pulse neural network, characterized in that: The device comprises a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the frosting area recognition method based on pulse neural network according to any one of claims 1-4 according to the instructions in the program code.

Images