A target detection method, device, equipment and computer-readable storage medium

Abstract

The present application discloses a target detection method, device, equipment and computer-readable storage medium, wherein the method comprises: in response to a target detection request, obtaining data to be detected; inputting the data to be detected into multiple target detection models at the same time, obtaining the detection results of each target detection model on the data to be detected, wherein the multiple target detection models are configured according to multiple subsets in the training data, the subset data in the same subset are the same target category, and the number of models of the target detection model is the same as the number of target categories of the training data; merging all the detection results to obtain the target detection results of the data to be detected. The technical problem that the existing target detection method has limited detection effect and low accuracy is solved.

Description

Technical Field

[0001] The present application relates to the field of target detection, and in particular to a target detection method, device, equipment and computer-readable storage medium. Background Art

[0002] With the development of computer technology, more and more attention has been paid to the research of computer vision, such as image classification, target detection and target tracking. Among them, target detection can distinguish the interesting targets from the uninteresting parts in the detection data, so as to carry out targeted processing on the targets, and has broad application prospects in the fields of security, military, transportation and medicine.

[0003] Taking intelligent driving as an example, target detection technology is the basis of intelligent driving, and its accuracy directly affects the safety and reliability of intelligent driving. However, the existing target detection methods have limited detection effects and low accuracy. Summary of the invention

[0004] The present application provides a target detection method, apparatus, device and computer-readable storage medium, which solve the technical problems of limited detection effect and low accuracy of existing target detection methods.

[0005] In view of this, the first aspect of the present application provides a target detection method, comprising:

[0006] Responding to a target detection request, obtaining data to be detected;

[0007] Inputting the data to be detected into multiple target detection models at the same time, and obtaining detection results of each target detection model on the data to be detected, wherein the multiple target detection models are configured according to multiple subsets in the training data, the subset data in the same subset are the same target category, and the number of models of the target detection models is the same as the number of target categories of the training data;

[0008] All the detection results are combined to obtain the target detection result of the data to be detected.

[0009] Optionally, the subset in the training data includes: an original subset and a newly added subset, and the target categories of the original subset and the newly added subset are different.

[0010] Optionally, the plurality of target detection models are configured according to a plurality of subsets in the training data, specifically including:

[0011] Obtaining subset data corresponding to the subset in the training data;

[0012] According to the subset data, the weight of the corresponding target detection model is updated to configure the corresponding target detection model.

[0013] Optionally, updating the weight of the corresponding target detection model according to the subset data to configure the corresponding target detection model specifically includes:

[0014] Inputting the subset data separately into each of the target detection models;

[0015] According to the subset data input into each target detection model, the weight of the target detection model is updated to configure the corresponding target detection model.

[0016] Optionally, when the data to be detected is image data to be detected, the method further includes:

[0017] According to the target detection result, the feature image of the image data to be detected is cropped to obtain the target features of the image data to be detected.

[0018] Optionally, the method further comprises:

[0019] According to the target features and the preset classification model, the target classification result of the image data to be detected is obtained.

[0020] Optionally, when the data to be detected is image data to be detected, and the image data to be detected is feature image data of an original image, the method further includes:

[0021] According to the target detection result, the image data to be detected is cropped to obtain target features.

[0022] Optionally, the method further comprises:

[0023] According to the target features and the preset classification model, the target classification result of the image data to be detected is obtained.

[0024] Optionally, the method further comprises:

[0025] According to the target feature, a frame regression is performed on the target detection box corresponding to the target feature in the image data to be detected.

[0026] Optionally, a plurality of said target detection models are integrated into one network.

[0027] A second aspect of the present application provides a target detection device, comprising:

[0028] A first acquisition unit, configured to acquire image data to be detected in response to a target detection request;

[0029] A second acquisition unit is used to input the data to be detected into multiple target detection models at the same time, and obtain detection results of each target detection model on the data to be detected, wherein the multiple target detection models are configured according to multiple subsets in the training data, the subset data in the same subset are the same target category, and the number of models of the target detection models is the same as the number of target categories of the training data;

[0030] The third acquisition unit is used to acquire the target detection result of the data to be detected according to all the detection results.

[0031] A third aspect of the present application provides a target detection device, the device comprising a processor and a memory;

[0032] The memory is used to store program code and transmit the program code to the processor;

[0033] The processor is used to execute the target detection method described in the first aspect according to the instructions in the program code.

[0034] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store program code, and the program code is used to execute the target detection method described in the first aspect.

[0035] It can be seen from the above technical methods that this application has the following advantages:

[0036] In the present application, in response to a target detection request, the data to be detected is obtained. Since multiple target detection models are configured according to multiple subsets in the training data, and a corresponding subset is a target category, that is, the target categories corresponding to the multiple target detection models obtained by configuration are different, when the data to be detected is input into multiple target detection models at the same time, the detection results output by each of the multiple target detection models can be obtained, and then all the detection results can be merged to obtain the target detection results of the data to be detected. The present application determines the target detection results based on the detection results of the detection data by multiple target detection models of different target categories. Compared with the existing method of performing target detection through a single target detection model, the accuracy is higher, thereby solving the technical problems of limited detection effect and low accuracy of existing target detection methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical methods in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0038] Figure 1 This is a flow chart of a first embodiment of a target detection method in an embodiment of the present application;

[0039] Figure 2 A schematic diagram of the process of configuring the target detection model in the embodiment of the present application;

[0040] Figure 3 This is a flow chart of a third embodiment of a target detection method in the embodiments of the present application;

[0041] Figure 4 This is a schematic diagram of a network structure in a fourth embodiment of a target detection method in an embodiment of the present application;

[0042] Figure 5 It is a structural schematic diagram of an embodiment of a target detection device in the embodiments of the present application. DETAILED DESCRIPTION

[0043] The embodiments of the present application provide a target detection method, apparatus, device and computer-readable storage medium, which solve the technical problems of limited detection effect and low accuracy of existing target detection methods.

[0044] In order to enable those skilled in the art to better understand the method of the present application, the technical method in the embodiment of the present application will be clearly and completely described below in conjunction with the drawings in the embodiment of the present application. Obviously, the described embodiment is only a part of the embodiment of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0045] For easier understanding, see Figure 1 , Figure 1 This is a flow chart of Example 1 of a target detection method in the embodiments of the present application.

[0046] A target detection method in this embodiment includes:

[0047] Step 101: In response to a target detection request, obtain data to be detected.

[0048] When performing target detection, first respond to the target detection request and obtain the data to be detected. Among them, the data type of the data to be detected can be multiple, for example, sound data, image data, video data, laser point cloud data, etc., and those skilled in the art can configure it according to their own detection needs. Accordingly, the collection or acquisition of data to be detected of different data types can be configured according to the type of data to be detected, for example, sound data is collected by a microphone, picture data is collected by a camera, etc., and those skilled in the art can set it according to needs, and will not be repeated here.

[0049] Step 102: input the data to be detected into multiple target detection models at the same time, and obtain the detection results of each target detection model on the data to be detected, wherein the multiple target detection models are configured according to multiple subsets in the training data, the subset data in the same subset are the same target category, and the number of models of the target detection model is the same as the number of target categories of the training data.

[0050] The multiple target detection models in this embodiment are configured according to multiple subsets in the training data, and the subset data in the same subset are of the same target category. One subset corresponds to one target category, and the target detection model configured according to the subset also corresponds to one target category. In this way, after the data to be detected is obtained, the data to be detected is simultaneously input into multiple target detection models, and the detection results of each target detection model on the data to be detected can be obtained.

[0051] It is understandable that the purpose of the target detection model is to distinguish the foreground (obstacle) and background in the data to be detected. The target category is the foreground of the target detection model. At this time, when the data to be detected is input into the target detection models of different target categories, each target detection model outputs the corresponding foreground. For example, the data to be detected is annotated with a lot of information, such as cars, carts, and bicycles. The data to be detected is simultaneously input into a car target detection model that only detects cars, a car target detection model that only detects cars, and a bicycle target detection model that only detects bicycles to distinguish the foreground and background. The detection result of the car target detection model is the car in the data to be detected, the detection result of the car target detection model is the car in the data to be detected, and the detection result of the bicycle target detection model is the bicycle in the data to be detected.

[0052] When the data to be detected is data of other data types, the target category of the target detection model can be set according to the data type and detection requirements of the data to be detected, which will not be described in detail here.

[0053] In one embodiment, multiple target detection models are integrated into one network, that is, each target detection model is a different sub-network in the network. When the network receives data to be detected, the data to be detected is input into different sub-networks of the network for target detection, and each sub-network outputs a corresponding detection result.

[0054] In another embodiment, multiple target detection models may also correspond to different networks at the same time. After the data to be detected is acquired, the data to be detected is simultaneously input into different networks to perform corresponding target detection.

[0055] It can be understood that a plurality means more than two, and the specific number can be set by those skilled in the art according to the analysis requirements of the detected data, which will not be limited or elaborated one by one here.

[0056] Step 103: Merge all detection results to obtain the target detection result of the data to be detected.

[0057] After obtaining the detection results output by each target detection model for the data to be detected, all the detection results can be combined to determine the target detection results of the data to be detected.

[0058] In one embodiment, a specific merging method for merging all detection results to obtain the target detection result of the data to be detected may be to perform a sum operation on all detection results. For example, when the detection result of the image data is a car, a car, and a bicycle, the target detection result of the image is a car + a car + a bicycle.

[0059] It is understandable that, in another embodiment, the test results may be merged by screening, that is, including:

[0060] Step 1031: Obtain the matching degree corresponding to each detection result, where the matching degree is the matching degree between the detection result and the target category of the target detection model.

[0061] Step 1032: After screening the test results according to the matching degree, a sum operation is performed on the screened test results.

[0062] That is, the detection results are screened according to the matching degree between each detection result and the target category of the target detection model. For example, the detection results of the image data are car, automobile, and bicycle, and the corresponding matching degrees are 95%, 85%, and 98%. At this time, the detection results with a matching degree less than 90% are eliminated, and the target detection result of the image is car + bicycle.

[0063] In the present application, in response to a target detection request, the data to be detected is obtained. Since multiple target detection models are configured according to multiple subsets in the training data, and a corresponding subset is a target category, that is, the target categories corresponding to the multiple target detection models obtained by configuration are different, when the data to be detected is input into multiple target detection models at the same time, the detection results output by each of the multiple target detection models can be obtained, and then all the detection results can be merged to obtain the target detection results of the data to be detected. The present application determines the target detection results based on the detection results of the detection data by multiple target detection models of different target categories. Compared with the existing method of performing target detection through a single target detection model, the accuracy is higher, thereby solving the technical problems of limited detection effect and low accuracy of existing target detection methods.

[0064] The above is Example 1 of a target detection method provided in an embodiment of the present application. The following is Example 2 of a target detection method provided in an embodiment of the present application. Based on Example 1, Example 2 provides a detailed description of the configuration process of multiple target detection models.

[0065] See also Figure 2 , Figure 2 Schematic diagram of the process of configuring the target detection model in the embodiment of the present application.

[0066] The multiple target detection models in this embodiment are configured according to multiple subsets in the training data, specifically including:

[0067] Step 201: Obtain subset data corresponding to a subset in training data.

[0068] Since a subset in the training data corresponds to a target category, and the number of models of the target detection model is the same as the number of targets in the training data, that is, one subset corresponds to one target detection model. Therefore, in this embodiment, the subset data corresponding to the subset in the training data is obtained. For example, the training data includes a small car subset, a large car subset, and a bicycle subset, and the corresponding subset data is the small car subset data, the large car subset data, and the bicycle subset data.

[0069] It is understandable that there are multiple subsets in the training data, and correspondingly there are multiple subset data in the subset, so obtaining the subset data can be randomly selecting a subset, and then randomly selecting a subset data in the subset. It should be noted that the above-mentioned method of obtaining the subset data can also be set to other methods by those skilled in the art, which will not be limited or elaborated here one by one.

[0070] Step 202: Update the weight of the corresponding target detection model according to the subset data to configure the corresponding target detection model.

[0071] In one implementation, updating the weight of the corresponding target detection model according to the subset data to configure the corresponding target detection model specifically includes:

[0072] Step 2021, input the subset data separately into each target detection model;

[0073] Step 2022: Update the weight of each target detection model according to the subset data input into the target detection model to configure the corresponding target detection model.

[0074] In one implementation, the subset data corresponding to each target detection model can be selected in turn and input into the corresponding target detection model to update the weight of the target detection model. It is understandable that in another implementation, the subset data is arbitrarily selected and input into each target detection model, and the target detection model determines whether the input subset data corresponds to the detection object. If yes, the weight of the target detection model is updated. If not, the weight of the current target detection model remains unchanged.

[0075] In this embodiment, the subset data in the same subset corresponds to a target category. After the subset data is obtained, the weight of the corresponding target detection model can be updated according to the subset data, thereby training the corresponding target detection model. For example, the weight of the small car target detection model is updated through the small car subset data, the weight of the large car target detection model is updated through the large car subset data, and the weight of the bicycle target detection model is updated through the bicycle subset data, and finally the configured small car target detection model, large car target detection model and bicycle target detection model are obtained.

[0076] It is understandable that, in one embodiment, multiple target detection models are integrated into one network, and after the network obtains a subset of data of a certain target category, the subset data is only input into the target detection model of the corresponding target category in the network, and the target detection model is configured, while other target detection models are ignored. In another embodiment, when multiple target detection models correspond to different networks, the subset data are input into different networks, and the weights of the networks are updated at the same time in the gradient update stage.

[0077] It is understandable that multiple subsets can constitute a mixed dataset. When configuring multiple target detection models, random sampling is used for the data in the mixed dataset to ensure that each subset and each target detection model are evenly trained.

[0078] For target detection, it is necessary to annotate all the object types of interest in each picture. If a certain type of object is not annotated in the picture, it will be treated as the background during training, which is harmful to the training of the model. The target detection model is the key to judging the foreground and background. In the prior art, due to different judgments of foreground and background in different data sets (inconsistent annotation categories), the foreground of one data set may be the background of another data set. If a new target category is to be detected, the target category that may exist in all the original data must be annotated. For example, the data in data set M has been annotated as: car, car and bicycle, and the corresponding target categories are also car, car and bicycle. At this time, if a new tricycle category is to be added, it is necessary to annotate all the tricycles that may exist in the data set, and then train the target detection model based on the annotated data to obtain a target detection model with target categories: car, car, bicycle and tricycle. The workload is large and cumbersome, which brings difficulties and inconsistency to the configuration of the target detection model.

[0079] In this embodiment, the subsets of training data include: the original subset and the newly added subset, and the target categories of the original subset and the newly added subset are different. The corresponding subsets are configured with the target detection model of the corresponding target category, that is, the new target detection model can be trained using the subset data in the newly added subset, and the original target detection model is obtained by training the original subset. In this way, when a new target category needs to be detected, only a new subset corresponding to the target category needs to be added, and then a new target detection model is configured according to the new subset.

[0080] For example, the original subsets in the training data A are the car subset, the large car subset and the bicycle subset, and the corresponding subset data are the car subset data, the large car subset data and the bicycle subset data. The target categories of the target detection model corresponding to each subset (that is, the target detection model corresponding to the training data A at this time) are cars, carts and bicycles. If you want to add the target category tricycle at this time, you only need to add a tricycle subset to the training data A, and then use the tricycle subset data in the tricycle subset to train the corresponding tricycle target detection model. At this time, the target categories of the target detection model corresponding to the training data A are also: cars, carts, bicycles and tricycles. The above process no longer re-annotates the subset data of the original subset in A (there may be a tricycle in A), and then trains the model through the A after the re-annotation. In this embodiment, when adding a new target category, it is only necessary to configure the corresponding target detection model using the subset data of the newly added target category. Compared with the existing training method, the training is simpler and more convenient.

[0081] The above is Example 2 of a target detection method provided in an embodiment of the present application. The following is Example 3 of a target detection method provided in an embodiment of the present application. Example 3 takes the data to be detected as the image data to be detected as an example to explain in detail the specific implementation of the target detection method.

[0082] See also Figure 3 , Figure 3 This is a flow chart of Example 3 of a target detection method in the embodiments of the present application.

[0083] A target detection method in this embodiment includes:

[0084] Step 301: In response to a target detection request, obtain image data to be detected.

[0085] It should be noted that step 301 is similar to the description of step 101 in the first embodiment. For details, please refer to the description of the above step 101, which will not be repeated here.

[0086] Step 302: input the data to be detected into multiple target detection models at the same time, and obtain the detection results of each target detection model on the data to be detected, wherein the multiple target detection models are configured according to multiple subsets in the training data, the subset data in the same subset are the same target category, and the number of target detection models is the same as the number of target categories of the training data.

[0087] It should be noted that step 302 is similar to the description of step 102 in the first embodiment. For details, please refer to the description of the above step 102, which will not be repeated here.

[0088] Step 303: merge all detection results to obtain the target detection result of the data to be detected.

[0089] It should be noted that step 303 is similar to the description of step 103 in the first embodiment. For details, please refer to the description of the above step 103, which will not be repeated here.

[0090] Step 304: According to the target detection result, the feature image of the image data to be detected is cropped to obtain the target features of the image data to be detected.

[0091] It can be understood that, in one embodiment, the image data to be detected is original image data. After obtaining the target detection result of the image data to be detected, the feature image of the image data to be detected is cropped according to the target detection result to obtain the target feature of the image data to be detected, which specifically includes:

[0092] Step 3041: extract features from the image data to be detected to obtain a feature image.

[0093] It is understandable that when the image data to be detected is raw image data, to obtain the target features, it is first necessary to extract features from the image data to be detected to obtain a feature image of the image data to be detected. Those skilled in the art can select a specific feature extraction method as needed, which will not be limited or elaborated herein.

[0094] Step 3042: According to the target detection result, the feature image is cropped to obtain the target feature.

[0095] It is understandable that after obtaining the target detection result, the target feature can be obtained by cropping the extracted feature image according to the target detection result.

[0096] It can be understood that, in another embodiment, the image data to be detected is a feature image of the original image, and according to the target detection result, the feature image of the image data to be detected is cropped to obtain the target feature of the image data to be detected, which specifically includes:

[0097] According to the target detection results, the image data to be detected is cropped to obtain the target features.

[0098] Since the data to be detected at this time is the feature image of the original image, when obtaining the target feature, there is no need to perform feature image processing on the image to be detected, and the target feature can be obtained by directly cropping the image data to be detected.

[0099] Step 305: Obtain target classification results of the image data to be detected according to the target features and the preset classification model.

[0100] After obtaining the target features, the target classification results of the image data to be detected can be obtained according to the target features and the preset classification model. It should be noted that the preset classification model can be a variety of model structures, as long as it can achieve target classification, and those skilled in the art can set it as needed, and no specific limitation is made here.

[0101] Step 306: Based on the target features, perform frame regression on the target detection box corresponding to the target features in the image data to be detected.

[0102] The purpose of bounding box regression is to use the mapping relationship to make the mapped target box of the candidate target box infinitely close to the real target box, so as to make the target positioning more accurate. Therefore, in this embodiment, according to the target features, the target detection box corresponding to the target features in the image data to be detected can also be regressed to more accurately locate the target.

[0103] It can be understood that the target detection method for the image data to be detected in this embodiment is only a schematic example. Those skilled in the art can adapt the above method according to usage requirements and apply it to laser point cloud data to be detected, sound data to be detected, etc.

[0104] In the present embodiment, in response to a target detection request, data to be detected is obtained. Since multiple target detection models are configured according to multiple subsets in the training data, and a corresponding subset is a target category, that is, the target categories corresponding to the multiple target detection models obtained by configuration are different, when the data to be detected is input into multiple target detection models at the same time, the detection results output by each of the multiple target detection models can be obtained, and then all the detection results can be merged to obtain the target detection results of the data to be detected. The present application determines the target detection results based on the detection results of the detection data by multiple target detection models of different target categories. Compared with the existing method of performing target detection through a single target detection model, the accuracy is higher, thereby solving the technical problems of limited detection effect and low accuracy of existing target detection methods.

[0105] The above is Example 3 of a target detection method provided in an embodiment of the present application, and the following is Example 4 of a target detection method provided in an embodiment of the present application.

[0106] First, the network model structure used in the target detection method in this embodiment is described. Figure 4 The network structure corresponding to the target detection method in this embodiment is shown, including: a backbone network, a target detection model, a region merging module, a feature region cropping module and a classification regression module. The following is an example of the image data to be detected to illustrate the various parts of the network model structure as follows:

[0107] The backbone network is used to extract features from the image data to be detected. The input image is a 4-order tensor with the size of each dimension being (total number of images, height, width, and three color channels). The backbone network uses a multi-layer convolutional neural network structure.

[0108] The target detection model is used to preliminarily estimate the target area (i.e., foreground) in the entire image and distinguish the target area from the background. The multiple target detection models in this embodiment are integrated on the backbone network. The network corresponding to the target detection model is a fully convolutional structure, which predicts whether the predefined anchor box is the background at each position.

[0109] The region merging module is used to merge the detection results of each target detection model.

[0110] The feature region cropping module is used to crop and scale the feature image extracted by the backbone network to obtain the target features based on the results obtained by the region merging module.

[0111] The classification and regression module is used to perform further classification and bounding box regression on the target features obtained by cropping in the previous step.

[0112] When using the above network model structure for target detection, it mainly includes two stages, training stage and application stage, which are:

[0113] First, assume that the different data sets (i.e., subsets) are D i , 1≤i≤N, and N independent target detection models RPN are constructed. i In the target detection model RPN i The back-end access area merging module is used to merge multiple RPN i The test results p i to merge.

[0114] Training phase: randomly select data set D i , and get a batch of data x from it i ,After the backbone network, enter the corresponding RPN i , ignoring the rest of the target detection models. The region merging module only receives one input, so it does not need to be merged. When the forward calculation of this step is completed and the gradient is updated, only the target detection model RPN used in this step is updated. i During the training process, since random sampling is used in each step, each data set D i and RPN i Will get even training.

[0115] Application phase: In the application phase, after the input data x passes through the backbone network, it needs to enter all target detection models to calculate the detection result proposals. The region merging module merges all RPNs. i Output.

[0116] In the present embodiment, in response to a target detection request, data to be detected is obtained. Since multiple target detection models are configured according to multiple subsets in the training data, and a corresponding subset is a target category, that is, the target categories corresponding to the multiple target detection models obtained by configuration are different, when the data to be detected is input into multiple target detection models at the same time, the detection results output by each of the multiple target detection models can be obtained, and then all the detection results can be merged to obtain the target detection results of the data to be detected. The present application determines the target detection results based on the detection results of the detection data by multiple target detection models of different target categories. Compared with the existing method of performing target detection through a single target detection model, the accuracy is higher, thereby solving the technical problems of limited detection effect and low accuracy of existing target detection methods.

[0117] The above is Example 4 of a target detection method provided in an embodiment of the present application, and the following is an embodiment of a target detection device provided in an embodiment of the present application.

[0118] See also Figure 5 , Figure 5 A schematic diagram of the structure of an embodiment of a target detection device in the embodiments of the present application.

[0119] A target detection device in this embodiment includes:

[0120] A first acquisition unit 501 is used to acquire image data to be detected in response to a target detection request;

[0121] A second acquisition unit 502 is used to input the data to be detected into multiple target detection models to obtain detection results of each target detection model on the data to be detected, wherein the target categories corresponding to the multiple target detection models are different;

[0122] The third acquisition unit 503 is used to acquire the target detection result of the data to be detected according to all the detection results.

[0123] Furthermore, when the target categories corresponding to the multiple target detection models are different, the third acquisition unit 503 is specifically used to merge all the detection results to obtain the target detection result of the data to be detected.

[0124] Furthermore, the subsets in the training data include: an original subset and a newly added subset, and the target categories of the original subset and the newly added subset are different.

[0125] Furthermore, multiple object detection models are configured based on multiple subsets of the training data, including:

[0126] Get the subset data corresponding to the subset in the training data;

[0127] According to the subset data, the weight of the corresponding target detection model is updated to configure the corresponding target detection model.

[0128] Further, according to the subset data, the weight of the corresponding target detection model is updated to configure the corresponding target detection model, specifically including:

[0129] Input the subset data separately into each target detection model;

[0130] According to the subset data input into each target detection model, the weight of the target detection model is updated to configure the corresponding target detection model.

[0131] Furthermore, when the data to be detected is image data to be detected, the target detection device in this embodiment further includes:

[0132] The cropping unit is used to crop the feature image of the image data to be detected according to the target detection result to obtain the target features of the image data to be detected.

[0133] Furthermore, the target detection device in this embodiment further includes:

[0134] The classification unit is used to obtain the target classification result of the image data to be detected according to the target features and the preset classification model.

[0135] Furthermore, the target detection device in this embodiment further includes:

[0136] The bounding box regression unit is used to perform bounding box regression on the target detection box corresponding to the target feature in the image data to be detected according to the target feature.

[0137] In the present application, in response to a target detection request, the data to be detected is obtained. Since multiple target detection models are configured according to multiple subsets in the training data, and a corresponding subset is a target category, that is, the target categories corresponding to the multiple target detection models obtained by configuration are different, when the data to be detected is input into multiple target detection models at the same time, the detection results output by each of the multiple target detection models can be obtained, and then all the detection results can be merged to obtain the target detection results of the data to be detected. The present application determines the target detection results based on the detection results of the detection data by multiple target detection models of different target categories. Compared with the existing method of performing target detection through a single target detection model, the accuracy is higher, thereby solving the technical problems of limited detection effect and low accuracy of existing target detection methods.

[0138] An embodiment of the present application also provides an embodiment of a target detection device. The detection device in this embodiment includes a processor and a memory; the memory is used to store program code and transfer the program code to the processor; the processor is used to execute the target detection method in the aforementioned embodiment according to the instructions in the program code.

[0139] An embodiment of the present application also provides an embodiment of a computer-readable storage medium. In this embodiment, the computer-readable storage medium is used to store program codes, and the program codes are used to execute the target detection method in the aforementioned embodiment.

[0140] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0141] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0142] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0143] In the several embodiments provided in the present application, it should be understood that the disclosed systems, commodity loading servers and methods can be implemented in other ways. For example, the commodity loading server embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, 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 through some interfaces, and the indirect coupling or communication connection of the commodity loading server or unit can be electrical, mechanical or other forms.

[0144] 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.

[0145] In addition, each functional unit in each embodiment of the present application 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.

[0146] 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 application is essentially 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 to enable 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 various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (full name in English: Read-Only Memory, English abbreviation: ROM), random access memory (full name in English: Random Access Memory, English abbreviation: RAM), disk or optical disk and other media that can store program codes.

[0147] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein 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 application.

Claims

1. A target detection method, characterized in that: include: In response to a target detection request, acquiring image data to be detected; Input the image data to be detected into multiple target detection models at the same time, and obtain the detection results of each target detection model on the image data to be detected, wherein the multiple target detection models are dynamically configured according to multiple subsets in the training data, the subset data in the same subset are the same target category, the number of models of the target detection model is the same as the number of target categories of the training data, and the subsets in the training data include: an original subset and a newly added subset, the target categories of the original subset and the newly added subset are different, and the newly added subset is used to dynamically configure a new target detection model when a new target category needs to be detected; Merging all the detection results to obtain the target detection result of the image data to be detected; The merging method of the test results is screening merging, and the screening merging process includes: Obtaining a matching degree corresponding to each of the detection results, the matching degree being a matching degree between the detection result and a target category of the target detection model; The detection results whose matching degree is less than a preset matching degree threshold are eliminated, and a sum operation is performed on the screened detection results to obtain the target detection result of the image data to be detected.

2. The target detection method according to claim 1, characterized in that: The target detection models are dynamically configured based on the subsets of training data, including: Obtaining subset data corresponding to the subset in the training data; According to the subset data, the weight of the corresponding target detection model is updated to configure the corresponding target detection model.

3. The target detection method according to claim 2, characterized in that: According to the subset data, updating the weight of the corresponding target detection model to configure the corresponding target detection model specifically includes: Inputting the subset data separately into each of the target detection models; According to the subset data input into each target detection model, the weight of the target detection model is updated to configure the corresponding target detection model.

4. The target detection method according to claim 1, characterized in that: The method further comprises: According to the target detection result, the feature image of the image data to be detected is cropped to obtain the target features of the image data to be detected.

5. The target detection method according to claim 4, characterized in that: The method further comprises: According to the target features and the preset classification model, the target classification result of the image data to be detected is obtained.

6. The target detection method according to claim 4, characterized in that: The method further comprises: According to the target feature, a frame regression is performed on the target detection box corresponding to the target feature in the image data to be detected.

7. The target detection method according to claim 1, characterized in that: A plurality of said target detection models are integrated into one network.

8. A target detection device, characterized in that: include: A first acquisition unit, configured to acquire image data to be detected in response to a target detection request; A second acquisition unit is used to input the image data to be detected into multiple target detection models at the same time, and obtain the detection results of each target detection model on the image data to be detected, wherein the multiple target detection models are dynamically configured according to multiple subsets in the training data, the subset data in the same subset are the same target category, the number of models of the target detection model is the same as the number of target categories of the training data, and the subsets in the training data include: an original subset and a newly added subset, the target categories of the original subset and the newly added subset are different, and the newly added subset is used to dynamically configure a new target detection model when a new target category needs to be detected; A third acquisition unit, used for acquiring a target detection result of the image data to be detected according to all the detection results; The merging method of the test results is screening merging, and the screening merging process includes: Obtaining a matching degree corresponding to each of the detection results, the matching degree being a matching degree between the detection result and a target category of the target detection model; The detection results whose matching degree is less than a preset matching degree threshold are eliminated, and a sum operation is performed on the screened detection results to obtain the target detection result of the image data to be detected.

9. A target detection device, 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 target detection method according to any one of claims 1 to 7 according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store program codes, and the program codes are used to execute the target detection method according to any one of claims 1 to 7.

Images