Method and system for classifying a target object based on photocurrent

Abstract

The application discloses a target object category identification method and system based on photocurrent, and relates to the technical field of target classification identification. The category identification method comprises the following steps: acquiring target infrared image data of a target object; determining a photocurrent data set of the target infrared image data according to the target infrared image data; determining a current data set according to the photocurrent data set and a preset weight set corresponding to the photocurrent data set; determining target category data of the target object according to the current data set, so as to determine the category of the target object according to the target category data. The category identification system provided by the application can determine the category of the target object through a full analog neural network, and can avoid using a digital-to-analog converter, thereby effectively reducing the power consumption of the category identification system.

Description

Technical Field

[0001] This application relates to the technical field of target classification and recognition, and more specifically, to a method and system for classifying target objects based on photocurrent. Background Technology

[0002] Infrared detection technology, as an important component of unmanned terminal detection systems, has shown application potential in the field of target classification and recognition. For example, infrared detection technology can be used for the identification of categories such as aircraft, automobiles, and ships.

[0003] Existing infrared detection systems rely on digital image processing technology, which involves the conversion between analog and digital signals. This requires the use of analog-to-digital converters (ADCs), significantly increasing the power consumption of the identification system.

[0004] The content of the background section is merely technology known to the public and does not necessarily represent existing technology in the field. Summary of the Invention

[0005] This application aims to provide a method and system for classifying target objects based on photocurrent, in order to solve the aforementioned technical problem of high system power consumption caused by the use of digital-to-analog converters.

[0006] According to one aspect of this application, a method for classifying a target object based on photocurrent is provided. The method includes: acquiring target infrared image data of the target object; determining a photocurrent dataset of the target infrared image data based on the target infrared image data; determining a current dataset based on the photocurrent dataset and a preset weight set corresponding to the photocurrent dataset; and determining target category data of the target object based on the current dataset, so as to determine the category of the target object based on the target category data.

[0007] According to some embodiments of this application, before the step of determining the current dataset based on the photocurrent dataset and the preset weight set corresponding to the photocurrent dataset, the category recognition method further includes: determining the preset weight set.

[0008] According to some embodiments of this application, the step of determining the current dataset based on the photocurrent dataset and the preset weight set corresponding to the photocurrent dataset includes: amplifying and converting the photocurrent dataset to obtain a first voltage dataset with a preset voltage range; determining current sub-data based on the first voltage dataset and the preset weight subset corresponding to the first voltage dataset; and traversing all preset weight subsets in the first voltage dataset and the preset weight set to obtain all current sub-data, so as to obtain the current dataset based on all current sub-data.

[0009] According to some embodiments of this application, the step of determining the target category data of the target object based on the current dataset includes: determining a second voltage dataset based on the current dataset; determining a differential voltage dataset based on the second voltage dataset; and determining the target category data based on the differential voltage dataset.

[0010] According to some embodiments of this application, the second voltage dataset includes at least one second voltage sub-dataset, and the second voltage sub-dataset includes two second voltage sub-datasets; the step of determining the differential voltage dataset based on the second voltage dataset includes: determining the differential voltage sub-dataset based on the difference between the two second voltage sub-datasets of the second voltage sub-dataset; traversing all the second voltage sub-datasets of the second voltage dataset to obtain all the differential voltage sub-datasets, so as to obtain the differential voltage dataset based on all the differential voltage sub-datasets.

[0011] According to some embodiments of this application, the step of determining the target category data based on the differential voltage dataset includes: comparing the differential voltage sub-data in the differential voltage dataset; and determining the differential voltage sub-data in the differential voltage dataset that meets the preset category conditions as the target category data.

[0012] According to one aspect of this application, a target object category recognition system driven by photocurrent is provided. The category recognition system includes an image acquisition module and a processing module. The image acquisition module acquires target infrared image data of the target object and determines a photocurrent dataset based on the target infrared image data. The input end of the processing module is connected to the output end of the image acquisition module. Based on the photocurrent dataset and a preset weight set corresponding to the photocurrent dataset, the processing module determines a current dataset and then determines the target object category data based on the current dataset, thereby determining the category of the target object based on the target category data.

[0013] According to some embodiments of this application, the category recognition system further includes a weight control module. The output of the weight control module is connected to the input of the processing module, and determines a preset weight set to output the preset weight set.

[0014] According to some embodiments of this application, the processing module includes an amplification unit and a first processing unit. The input of the amplification unit is connected to the output of the image acquisition module, and amplifies and converts the photocurrent dataset to obtain a first voltage dataset within a preset voltage range. The input of the first processing unit is connected to the output of the amplification unit, and determines current sub-data based on the first voltage dataset and a preset weight subset corresponding to the first voltage dataset. It then iterates through all preset weight subsets in the first voltage dataset and the preset weight set to obtain all current sub-data, thereby obtaining a current dataset based on all current sub-data.

[0015] According to some embodiments of this application, the processing module further includes a current-to-voltage conversion unit, a differential circuit unit, and a second processing unit. The input terminal of the current-to-voltage conversion unit is connected to the output terminal of the first processing unit, and a second voltage dataset is determined based on the current dataset. The input terminal of the differential circuit unit is connected to the output terminal of the current-to-voltage conversion unit, and a differential voltage dataset is determined based on the second voltage dataset. The input terminal of the second processing unit is connected to the output terminal of the differential circuit unit, and target category data is determined based on the differential voltage dataset.

[0016] Beneficial effects

[0017] This application can determine the photocurrent dataset from the target infrared image data of the target object. This application can determine the current dataset using the photocurrent dataset and a preset weight set corresponding to the photocurrent dataset. This application can determine the target category data of the target object using the current dataset, and then determine the category of the target object based on the target category data.

[0018] The category recognition system provided in this application determines the category of the target object through a fully analog neural network, which can avoid the use of a digital-to-analog converter and thus effectively reduce the power consumption of the category recognition system. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A schematic diagram of a category identification system according to an embodiment of this application is shown;

[0021] Figure 2 This invention provides another schematic diagram of the structure of a category identification system according to an embodiment of the present application.

[0022] Figure 3 This invention provides another schematic diagram of the structure of a category identification system according to an embodiment of the present application.

[0023] Figure 4 This illustration shows a flowchart of a category identification method 1000 according to an embodiment of the present application;

[0024] Figure 5 This illustrates another flowchart of a category identification method 1000 according to an embodiment of the present application;

[0025] Figure 6 A flowchart illustrating step S130 according to an embodiment of this application is shown.

[0026] Figure 7 A flowchart illustrating step S140 according to an embodiment of this application is shown.

[0027] Figure 8 A flowchart illustrating step S142 according to an embodiment of this application is shown;

[0028] Figure 9 A flowchart illustrating step S143 according to an embodiment of this application is shown.

[0029] Explanation of reference numerals in the attached figures:

[0030] Category recognition system 200.

[0031] Image acquisition module 210; processing module 220; weight control module 230.

[0032] Amplification unit 221; first processing unit 222; current-to-voltage conversion unit 223; differential circuit unit 224; second processing unit 225.

[0033] Near-infrared detector 211,

[0034] Photocurrent amplifier 2211.

[0035] Memristor subunit 2221.

[0036] Current-to-voltage converter 2231.

[0037] Differential circuit 2241.

[0038] Weight writing module 231; weight switch group 232;

[0039] Weight switch 2321. Detailed Implementation

[0040] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided so that this application will be thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0041] The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of these specific details, or other methods, components, materials, devices, etc. In these cases, well-known structures, methods, devices, implementations, materials, or operations will not be shown or described in detail.

[0042] Furthermore, the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0043] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, rather than to describe a specific order.

[0044] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0045] According to one aspect of this application, a photocurrent-driven target object category identification system 200 is provided. See also Figure 1 The category recognition system 200 may include an image acquisition module 210 and a processing module 220. See also... Figure 2 The category recognition system 200 may also include a weight control module 230. See also Figure 3 The processing module 220 includes an amplification unit 221, a first processing unit 222, a current-to-voltage conversion unit 223, a differential circuit unit 224, and a second processing unit 225.

[0046] The following is combined Figure 1 , Figure 2 and Figure 3 This application describes a target object category identification method 1000 based on photocurrent. The category identification method 1000 can be executed by the category identification system 200.

[0047] See Figure 4 The category identification method 1000 may include steps S110-S140.

[0048] In step S110, the image acquisition module 210 acquires the infrared image data of the target object.

[0049] According to the example embodiment, the target object can be a target object of a category to be identified. The target infrared image data can be the infrared image data of the target object.

[0050] The image acquisition module 210 can acquire target infrared image data through infrared image sensors such as near-infrared image sensors, mid-infrared image sensors, and far-infrared image sensors.

[0051] The type of target object can be selected according to the identification requirements, and this application does not impose any restrictions. For example, the target object can be one of the three types of targets to be identified: an airplane, an oil drum, and a ship.

[0052] The image acquisition module 210 is an i×j array near-infrared detector 211. Here, i is the number of rows in the i×j array near-infrared detector 211, and j is the number of columns in the i×j array near-infrared detector 211. The target infrared image data acquired by the image acquisition module 210 is an infrared image of the i×j array. The number of pixels in the target infrared image data is m, and the value of m is the product of i and j.

[0053] For example, see Figure 3 The image acquisition module 210 is an 8×8 array near-infrared detector 211, which acquires 8×8 array infrared image data of the target object. The 8×8 array infrared image data of the target object includes 64 pixels.

[0054] In step S120, the image acquisition module 210 determines the photocurrent dataset of the target infrared image data based on the target infrared image data.

[0055] According to an example embodiment, one pixel of the target infrared image data is converted into one photocurrent sub-data. All photocurrent sub-data form a photocurrent dataset. The photocurrent dataset can be a collection of photocurrent sub-data converted from pixels of the target infrared image data.

[0056] The image acquisition module 210 can convert each pixel data in the target infrared image data into corresponding photocurrent sub-data using near-infrared image sensors, mid-infrared image sensors, far-infrared image sensors, etc., and form a photocurrent dataset based on the photocurrent sub-data. The photocurrent dataset can be a 1×m matrix.

[0057] For example, the image acquisition module 210 can convert 64 pixel data from the acquired 8×8 array of target infrared image data into 64 photocurrent sub-data, with a one-to-one correspondence between the 64 pixel data and the 64 photocurrent sub-data. The image acquisition module 210 can then form a photocurrent dataset from the 64 photocurrent sub-data and output the photocurrent dataset. The photocurrent dataset can be a 1×64 matrix.

[0058] In step S130, the processing module 220 determines the current dataset based on the photocurrent dataset and the preset weight set corresponding to the photocurrent dataset.

[0059] According to the example embodiment, the preset weight set can be a set of preset weights corresponding to the photocurrent dataset. The preset weight set can be an m×n matrix, where n is twice the number of target categories. The number of target categories is the number of types of target objects. The weights in the preset weight set correspond to photocurrent sub-data.

[0060] For example, if the target object can be one of three types of objects to be identified, such as an airplane, an oil drum, or a ship, then n is 6.

[0061] The number of target categories can be set according to the application scenario of the identification, and this application does not impose any restrictions. For example, the target objects can also be any 5 or 10 types of objects, in which case n would correspond to 10 or 20 respectively.

[0062] Photocurrent quantum data is obtained by multiplying photocurrent quantum data with its corresponding preset weights. All current quantum data form a set of current quantum data, i.e., a current dataset. The current dataset can be a 1×n matrix.

[0063] The input terminal of the processing module 220 is connected to the output terminal of the image acquisition module 210. The processing module 220 can determine the current dataset based on the photocurrent dataset and a preset weight set through a single fully connected neural network.

[0064] For example, the processing module 220 can determine the current dataset (1×6 matrix) based on the photocurrent dataset (1×64 matrix) and the preset weight set (64×6 matrix) through a fully connected network.

[0065] In step S140, the processing module 220 determines the target category data of the target object based on the current dataset, so as to determine the category of the target object based on the target category data.

[0066] According to the example embodiment, the target category data can be the category data of the target object. The processing module 220 can convert the current dataset into a voltage dataset, and perform differential processing on two adjacent voltage data in the voltage dataset to obtain a differential voltage dataset. The processing module 220 can compare the various differential voltage sub-data in the differential voltage dataset, and the differential voltage sub-data is the probability distribution of the predicted target category data.

[0067] Processing module 220 selects the maximum value in the differential voltage sub-data as the target category data, and then determines the category of the target object based on the probability distribution of the target category data. The target category data represents the maximum probability of the predicted target object category.

[0068] Through the above embodiments, this application can determine the photocurrent dataset of the target infrared image data using the target infrared image data of the target object. This application can determine the current dataset using the photocurrent dataset and a preset weight set corresponding to the photocurrent dataset. This application can determine the target category data of the target object using the current dataset, and then determine the category of the target object based on the target category data.

[0069] The category recognition system provided in this application determines the category of the target object through a fully analog neural network, which can avoid the use of a digital-to-analog converter and thus effectively reduce the power consumption of the category recognition system.

[0070] Optionally, see Figure 5 After step S120 and before step S130, the category identification method 1000 may further include step S120a.

[0071] The weight control module 230 executes step S120a. Step S120a is the step of determining the preset weight set.

[0072] According to the example embodiment, the output terminal of the weight control module 230 is connected to the input terminal of the processing module 220, and the weight control module 230 determines a preset weight set to output the preset weight set.

[0073] See Figure 3 The weight control module 230 may include a weight writing module 231 and a weight switch group 232.

[0074] According to the example embodiment, the weight switch group 232 may include a plurality of weight switches 2321. The number of weight switches 2321 can be the sum of m and n. The number of row weight switches in the weight switch group 232 is m, and the number of column weight switches in the weight switch group 232 is n.

[0075] The weight writing module 231 can control the writing of a preset weight set by controlling the on and off states of the weight switch 2321. When the weight switch 2321 is selectively on, the weight writing module 231 writes the preset weight set to the processing module 220. When the weight switch 2321 is off, the weight writing module 231 stops writing the preset weight set.

[0076] Optionally, see Figure 6 Step S130 may include steps S131-S133.

[0077] In step S131, the amplification unit 221 amplifies and converts the photocurrent dataset to obtain a first voltage dataset within a preset voltage range.

[0078] According to the example embodiment, the input terminal of the amplification unit 221 is connected to the output terminal of the image acquisition module 210. The amplification unit 221 can be a 1×m array photocurrent amplifier 2211. The amplification unit 221 can be based on a transimpedance amplifier circuit structure, using capacitor compensation to stabilize the transimpedance amplifier circuit, and taking into account parameters such as gain, bandwidth, noise, stability and power consumption, the circuit design with the lowest power consumption is selected as the target circuit design for the amplification unit 221.

[0079] For example, the transimpedance amplifier circuit structure can be a forward amplification structure or a reverse amplification structure.

[0080] The first voltage sub-data is obtained after amplification and conversion of the photocurrent sub-data. All the first voltage sub-data forms the first voltage dataset. The first voltage dataset can be a collection of the first voltage sub-data obtained after amplification and conversion of the photocurrent sub-data. The first voltage dataset includes at least one first voltage sub-data. The first voltage dataset can be a 1×m matrix.

[0081] For example, the first voltage dataset can be a 1×64 matrix.

[0082] The preset voltage range can be the range of voltage values ​​that the input terminal of the first processing unit 222 can withstand. For example, the preset voltage range can be the voltage value that the input terminal of the memristor analog computing circuit can withstand.

[0083] For example, the peak current signal of the photocurrent quantum data is approximately 100 nA, the input of the memristor analog computing circuit can withstand a voltage of 0.1 V, and the gain of the amplifier unit 221, which employs a transimpedance amplifier circuit structure, is 10. 6 According to the noise-free test calculation method for 8-bit infrared images, the noise of the amplification unit 221 must be less than 390uV.

[0084] In step S132, the first processing unit 222 determines the current sub-data based on the first voltage dataset and the preset weight subset corresponding to the first voltage dataset.

[0085] According to the example embodiment, the input terminal of the first processing unit 222 is connected to the output terminal of the amplification unit 221. The first processing unit 222 can be an m×n array of memristor sub-units 2221, and the number of memristor sub-units 2221 can be the product of m and n.

[0086] The preset weight set can be a pre-defined set of weights corresponding to the first voltage dataset. The preset weight set can be an m×n matrix.

[0087] The preset weight subset can be a subset of the weights corresponding to the current sub-data in the preset weight set. The preset weight subset can be an m×1 matrix. For example, the preset weight subset can be a 64×1 matrix.

[0088] The current sub-data can be the product of the first voltage dataset and the preset weighted subset, that is, the current sub-data can be the product of the first voltage dataset (a 1×m matrix) and the preset weighted subset (an m×1 matrix).

[0089] For example, the first processing unit 222 can determine the current sub-data according to the following formula:

[0090] ;

[0091] in, For one current quantum data; , ... () is the first voltage dataset. , ,……and All of these are the first voltage sub-data points from the first voltage dataset. , ... ) is a preset weighted subset corresponding to the first voltage dataset. for( , ... ) and The corresponding preset weight value. for( , ... ) and The corresponding preset weight value. for( , ... ) and The corresponding preset weight value.

[0092] In step S133, the first processing unit 222 traverses all preset weight subsets in the first voltage dataset and the preset weight set to obtain all current sub-data, so as to obtain the current dataset based on all current sub-data.

[0093] According to the example embodiment, the first processing unit 222 can perform matrix multiplication and addition operations according to Kirchhoff's current law. The first processing unit 222 can adopt a cross-array structure, where the row data (i.e., the input of the first processing unit 222) is a first voltage dataset. The column data (i.e., the output of the first processing unit 222) is a current dataset. The cross-array structure can achieve ±1 binary weighting by subtracting the weights (0 or 1) of two columns of memristors.

[0094] The first processing unit 222 can perform a product operation on the first voltage dataset and all preset weighted subsets to obtain all current sub-data. The first processing unit 222 obtains the current dataset based on all the current sub-data.

[0095] For example, the first processing unit 222 can determine the current dataset according to the following formula:

[0096] ;

[0097] in, For another current quantum data; ( , ... () is another preset weight subset corresponding to the first voltage dataset.

[0098] for( , ... ) and The corresponding preset weight value. for( , ... ) and The corresponding preset weight value. for( , ... ) and The corresponding preset weight value.

[0099] The current dataset can be a 1×n matrix, i.e. ( , ... For example, the current dataset can be a 1×6 matrix. The preset weight set can be:

[0100] ;

[0101] Each preset weight in the preset weight set can be either 0 or 1. These preset weights can be obtained by training on the category data of the images of the object being pre-trained. The category of the pre-trained object can be the same as the category of the target object.

[0102] The weight writing module 231 can be an MCU controller (Memory Controller Unit). The MCU controller can selectively control the weight switch group 232 by connecting it to high-level or low-level signals output by the MCU controller to control the values ​​of each preset weight. For example, if the MCU controller selects to close the weight switch 2321 in the first row and first column and connects it to a high level, the preset weight value corresponding to this high-level signal is 1. If the MCU controller selects to close the weight switch 2321 in the first row and first column and connects it to a low level, the preset weight value corresponding to this low-level signal is 0. The number of switch control signals of the MCU controller can be the sum of m and n. Furthermore, the high and low level signals selectively controlled and output by the MCU controller correspond one-to-one with the preset weight values.

[0103] Through the above embodiments, this application can obtain a first voltage dataset with a preset voltage range by amplifying and converting the photocurrent dataset. This application can determine current sub-data using the first voltage dataset and a preset weight subset corresponding to the first voltage dataset. This application can obtain all current sub-data by traversing the first voltage dataset and all preset weight subsets in the preset weight set, thereby obtaining the current dataset.

[0104] The category recognition system provided in this application can be driven by photocurrent and a preset weight set can be determined by binary weights, which makes the structure of the category recognition system of this application simple and the power consumption low.

[0105] Optionally, see Figure 7 Step S140 may include steps S141-S143.

[0106] In step S141, the current-to-voltage conversion unit 223 determines the second voltage dataset based on the current dataset.

[0107] According to the example embodiment, the input terminal of the current-to-voltage conversion unit 223 is connected to the output terminal of the first processing unit 222. The current-to-voltage conversion unit 223 can be a 1×n array of current-to-voltage converters 2231.

[0108] For example, the current-to-voltage conversion unit 223 can be a 1×6 array of current-to-voltage converters 2231.

[0109] The second voltage dataset can be the voltage set formed after converting the current sub-data into second voltage sub-data. The second voltage dataset can be a 1×n matrix.

[0110] For example, the second voltage dataset can be a 1×6 matrix.

[0111] The current-to-voltage conversion unit 223 converts all current sub-data into corresponding second voltage sub-data, and all the second voltage sub-data forms a second voltage dataset.

[0112] In step S142, the differential circuit unit 224 determines the differential voltage dataset based on the second voltage dataset.

[0113] According to the example embodiment, the input terminal of the differential circuit unit 224 is connected to the output terminal of the current-to-voltage conversion unit 223. The differential circuit unit 224 can be a 1×(n / 2) array of differential circuits 2241. For example, the differential circuit unit 224 can be a 1×3 array of differential circuits 2241.

[0114] Differential voltage sub-data is obtained by the difference between two adjacent second voltage sub-data. All differential voltage sub-data forms a differential voltage dataset. The differential voltage dataset can be a collection of differential voltage sub-data obtained by the difference between two adjacent second voltage sub-data. The differential voltage dataset can be a 1×(n / 2) matrix.

[0115] For example, the differential voltage dataset can be a 1×3 matrix.

[0116] See Figure 3 The two input terminals of the differential circuit 2241 are respectively connected to the output terminals of the two current-to-voltage converters 2231, and the two current-to-voltage converters 2231 are adjacent to each other.

[0117] The differential circuit 2241 subtracts two adjacent second voltage sub-data points from the second voltage data to obtain differential voltage sub-data points. All differential voltage sub-data points form a differential voltage dataset.

[0118] Optionally, the second voltage dataset includes at least one second voltage sub-dataset, and the second voltage sub-dataset includes two second voltage sub-datasets.

[0119] See Figure 8 Step S142 may include steps S1421-S1422.

[0120] In step S1421, the differential circuit unit 224 determines the differential voltage sub-data based on the difference between the two second voltage sub-data in the second voltage sub-data set.

[0121] According to the example embodiment, the second voltage subset is a set formed by two adjacent second voltage subsets in the second voltage dataset. The second voltage subset can be a 1×2 matrix. Two adjacent second voltage subsets are in the second voltage subset, and the index of the second second voltage subset is even.

[0122] For example, the first and second sub-data points of the second voltage dataset form a second voltage sub-data set. The third and fourth sub-data points of the second voltage dataset form a second voltage sub-data set. And so on, the (n-1)th and nth sub-data points of the second voltage dataset form a second voltage sub-data set.

[0123] Differential voltage subdata can be the difference between two second voltage subdata in a second voltage subset.

[0124] One of the differential circuit units 224, differential circuit 2241, can determine the differential voltage sub-data based on the difference between two second voltage sub-data in the second voltage sub-data set.

[0125] In step S1422, the differential circuit unit 224 traverses all the second voltage sub-data sets of the second voltage dataset to obtain all the differential voltage sub-data sets, so as to obtain the differential voltage dataset based on all the differential voltage sub-data sets.

[0126] According to the example embodiment, the differential circuit 2241 can correspond one-to-one with the second voltage subset.

[0127] Each differential circuit 2241 in the differential circuit unit 224 can determine the corresponding differential voltage sub-data based on the difference between two second voltage sub-data in the second voltage subset. All the differential voltage sub-data form a differential voltage dataset.

[0128] In step S143, the second processing unit 225 determines the target category data based on the differential voltage dataset.

[0129] According to the example embodiment, the input terminal of the second processing unit 225 is connected to the output terminal of the differential circuit unit 224. The second processing unit 225 can be a processor with data processing capabilities.

[0130] The input terminal of the second processing unit 225 can be connected to the output terminal of each differential circuit 2241. The second processing unit 225 can receive a differential voltage dataset. The second processing unit 225 can compare all the differential voltage sub-data in the differential voltage dataset and select the differential voltage sub-data with the maximum value as the target category data. The maximum value can be a positive maximum value or a negative maximum value.

[0131] Optionally, see Figure 9 Step S143 may include steps S1431-S1432.

[0132] In step S1431, the second processing unit 225 compares the differential voltage sub-data in the differential voltage dataset.

[0133] According to the example embodiment, the values ​​of the differential voltage sub-data can be positive, negative, or zero. Each differential voltage sub-data corresponds to the probability of the predicted target category data.

[0134] When the transimpedance amplifier circuit has a forward amplification structure, the second processing unit 225 can directly compare the differential voltage sub-data.

[0135] When the transimpedance amplifier circuit has an inverting amplifier structure, the second processing unit 225 can add a negative sign before the differential voltage sub-data and then compare them.

[0136] In step S1432, the second processing unit 225 determines the differential voltage data in the differential voltage dataset that meets the preset category conditions as the target classification data.

[0137] According to the example embodiment, the preset category condition rule can be the maximum positive value of the differential voltage sub-data. The preset category condition rule can also be the maximum negative value of the differential voltage sub-data.

[0138] When the transimpedance amplifier circuit has a forward amplification structure, the preset category condition rule can be the maximum positive value of the differential voltage sub-data.

[0139] For example, if the transimpedance amplifier circuit structure can be a forward amplification structure, and the values ​​of the three differential voltage sub-data in the differential voltage dataset are -1, 1, and -3 respectively, then the second processing unit 225 determines that the category corresponding to the differential voltage sub-data with a value of 1 is the target category data.

[0140] When the transimpedance amplifier circuit structure is an inverted amplification structure, the preset category condition rule can be the maximum negative value of the differential voltage sub-data.

[0141] For example, if the transimpedance amplifier circuit structure can be an inverted amplifier structure, and the values ​​of the three differential voltage sub-data points in the differential voltage dataset are -1, 1, and -3, then the second processing unit 225 determines that the category corresponding to the differential voltage sub-data point with the value of -3 is the target category data.

[0142] Through the above embodiments, this application can determine a second voltage dataset through a current dataset, determine a differential voltage dataset through a second voltage dataset, and determine target category data through a differential voltage dataset.

[0143] The processing module of the category recognition system provided in this application has a simple circuit structure, and the processing module operates by being driven by photocurrent, which can avoid the circuit in the image acquisition module that converts photocurrent into other digital signal forms, thereby further reducing the power consumption of the category recognition system.

[0144] According to the example embodiment, the number of target categories in the category recognition system provided in this application is backward compatible. For example, if the category recognition system has a preset number of 10 target categories, then the category recognition system can recognize any number of target categories from 1 to 10.

[0145] Finally, it should be noted that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions of the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for classifying target objects based on photocurrent, characterized in that, The category identification method includes: Acquire the infrared image data of the target object; Each pixel of the target infrared image data is converted into a photocurrent sub-data. All the photocurrent sub-data form a photocurrent dataset, which is a 1×m matrix, where m is the number of pixels in the target infrared image. Based on the photocurrent dataset and the preset weight set corresponding to the photocurrent dataset, a current dataset is determined, including: Photocurrent subdata is obtained by multiplying the photocurrent subdata with the preset weights corresponding to the photocurrent subdata. All the current subdata form a current dataset. The preset weight set is an m×n matrix, where n is twice the number of target categories. Each preset weight value in the preset weight set is trained based on the category data of the pre-trained object image. The current dataset is a 1×n matrix. The target category data of the target object is determined based on the current dataset, and the category of the target object is determined based on the target category data; The step of determining the target category data of the target object based on the current dataset includes: A second voltage dataset is determined based on the current dataset, wherein the second voltage dataset is a voltage set formed after the current sub-data is converted into the second voltage sub-data, and the second voltage dataset is a 1×n matrix; Determine the differential voltage dataset based on the second voltage dataset; The target category data is determined based on the differential voltage dataset, including: Compare the differential voltage sub-data in the differential voltage dataset; The differential voltage data that meets the preset category conditions in the differential voltage dataset are determined as the target category data, wherein the preset category conditions are the maximum positive value or the maximum negative value of the differential voltage sub-data. The step of determining the differential voltage dataset based on the second voltage dataset includes: Differential voltage sub-data is obtained based on the difference between two adjacent second voltage sub-data. All differential voltage sub-data forms a differential voltage dataset, which is a 1×(n / 2) matrix.

2. The category identification method according to claim 1, characterized in that, The step of determining the current dataset based on the photocurrent dataset and the preset weight set corresponding to the photocurrent dataset includes: The photocurrent dataset is amplified and converted to obtain a first voltage dataset within a preset voltage range; Based on the first voltage dataset and the preset weight subset corresponding to the first voltage dataset, the current sub-data is determined; Traverse the first voltage dataset and all preset weight subsets in the preset weight set to obtain all current sub-data, so as to obtain the current dataset based on all the current sub-data.

3. A target object category identification system driven by photocurrent, characterized in that, The method for classifying target objects based on photocurrent as described in any one of claims 1-2 includes: The image acquisition module acquires the target infrared image data of the target object, and determines the photocurrent dataset of the target infrared image data based on the target infrared image data; The processing module has its input end connected to the output end of the image acquisition module. Based on the photocurrent dataset and the preset weight set corresponding to the photocurrent dataset, it determines the current dataset, determines the target category data of the target object based on the current dataset, and determines the category of the target object based on the target category data.

4. The category recognition system according to claim 3, characterized in that, The category recognition system also includes: The weight control module has its output end connected to the input end of the processing module to determine the preset weight set and output the preset weight set.

5. The category recognition system according to claim 3, characterized in that, The processing module includes: The amplification unit, with its input end connected to the output end of the image acquisition module, amplifies and converts the photocurrent dataset to obtain a first voltage dataset within a preset voltage range. The first processing unit has its input end connected to the output end of the amplification unit. Based on the first voltage dataset and the preset weight subset corresponding to the first voltage dataset, it determines the current sub-data. It then traverses all preset weight subsets in the first voltage dataset and the preset weight set to obtain all the current sub-data, so as to obtain the current dataset based on all the current sub-data.

6. The category recognition system according to claim 5, characterized in that, The processing module further includes: A current-to-voltage conversion unit, with its input terminal connected to the output terminal of the first processing unit, determines a second voltage dataset based on the current dataset. The differential circuit unit has its input terminal connected to the output terminal of the current-to-voltage conversion unit, and determines the differential voltage dataset based on the second voltage data. The second processing unit has its input terminal connected to the output terminal of the differential circuit unit, and determines the target category data based on the differential voltage dataset.

Images