Target detection method, system and device based on big data empowerment and model flow

Abstract

The present invention belongs to the field of computer vision and pattern recognition, and specifically relates to a target detection method, system and device based on big data empowerment and model flow, which aims to solve the problem that the existing target detection model is limited by training data, resulting in low model performance and poor reusability in different application scenarios. The present invention includes: integrating all public target detection data sets, and building a model sampling space based on any model; completing a dynamic supernet covering various operating requirements in one training; in the current scene, initializing the model and screening the sub-model through the semantic information vector of the category; pre-training the sub-model through the current scene data, and the target detection model finally obtained can perform target detection of the image to be detected in the current scene. After constructing a flexible dynamic supernet, the present invention uses a small amount of labeled data in the current scene for fast fine-tuning to obtain a target detection model with excellent performance in the current scene.

Description

Background Art

[0002] Object detection is an important and challenging computer vision task, which has a wide range of applications in security monitoring, intelligent video analysis, autonomous driving and other fields. In various application scenarios, it is too costly to make the dataset reach the same level of COCO annotation. Generally, only a small amount of low-cost annotated data can be obtained, and the limited data makes model training very difficult, making it difficult to maximize the performance of the model. At the same time, in various application scenarios, the available hardware resources vary significantly, and the models that can be deployed are also different. In new scenarios, new models need to be selected from scratch based on human expert experience for retraining. The model has poor reusability and wastes hardware resources. Summary of the invention

[0003] In order to solve the above problems in the prior art, namely, the existing target detection model is limited by insufficient training data, resulting in low model performance, and the poor reusability of the model due to changes in hardware requirements in different application scenarios, the present invention provides a target detection method based on big data empowerment and model flow, which includes:

[0004] Step S10, obtaining the target detection image of the current scene, and converting the category information of the image into a semantic information vector through the Word2vector model;

[0005] Step S20, performing cosine similarity matching between the semantic information vector and the semantic information vectors of image category information of all public object detection datasets;

[0006] Step S30, initializing the object detection model of the current scene using the classification fully connected weights of the category with the highest matching value in the supernet with variable width and depth;

[0007] Step S40, pre-specify the number of floating-point operations and the amount of parameters of the model, traverse the model sampling space one by one, and obtain K target detection sub-models of the current scene;

[0008] Step S50, pre-training the K object detection sub-models using the training images of the current scene respectively, and taking the pre-trained sub-model with the largest mAP value as the final object detection model for the current scene;

[0009] Step S60, performing target detection on the target detection image of the current scene using the final target detection model of the current scene to obtain a target detection result.

[0010] In some preferred embodiments, the model sampling space is obtained by:

[0011] Step A10, normalizing the images in all the public object detection datasets obtained, and converting the category information corresponding to the images into semantic information vectors through the Word2vector model;

[0012] Step A20, calculating the cosine similarity between the semantic information vectors of each category, and merging the categories whose cosine similarity is greater than a preset threshold, and performing label remapping;

[0013] Step A30, select any deep learning target detection model, and set the width and depth sampling space of the model for the feature extraction backbone network of the model to obtain the model sampling space.

[0014] In some preferred embodiments, the construction and training method of the variable width and depth supernet is as follows:

[0015] Step B10, dividing the image set after label remapping into batches of a set size, extracting features of any batch of images through the feature extraction backbone network, and randomly sampling a model in the model sampling space;

[0016] Step B20, input the features of the current batch of images into the classification and bounding box regression branches of the corresponding random sampling model for forward propagation, and calculate the global loss of the model;

[0017] Step B30, reducing the global loss by back propagation method and stochastic gradient descent method to update model parameters, iterating training until the set training end condition is reached, and obtaining a super network with variable width and depth.

[0018] In some preferred embodiments, before dividing the image set after label remapping into batches of a set size in step B10, an image set expansion step is further provided, and the method is as follows:

[0019] The label-remapped images are subjected to random multi-scale scaling and multi-angle flipping operations to obtain an expanded image set.

[0020] In some preferred embodiments, the global loss is expressed as:

[0021] L all =λL rcnn +L rpn

[0022] Among them, L rcnn represents the classification and bounding box regression loss of the model, L rpn The region proposal network part of the model has a loss, and λ is a balancing factor used to balance the two losses.

[0023] In some preferred embodiments, the classification and bounding box regression losses of the model are expressed as:

[0024]

[0025] Among them, k represents the sequence number of the prediction box, p ki represents the predicted probability that the k-th prediction box is predicted as the i-th category, represents the actual probability that the k-th prediction box corresponds to the i-th class label, L cls is the cross entropy loss, N cls Represents the total number of categories included in the data set; t ki represents the predicted coordinates of the kth prediction box, represents the real coordinates corresponding to the k-th prediction box, L reg To smooth the L1 loss, N reg represents the total number of prediction boxes; γ is the balance factor used to balance the two losses.

[0026] In some preferred embodiments, the region proposal network part of the model is lost, which is expressed as:

[0027]

[0028] Among them, k represents the sequence number of the prediction box, p i Represents the predicted probability of whether the k-th prediction box contains an object, represents the actual probability of whether the k-th prediction box contains an object, L cls is the binary cross entropy loss, K represents the number of all predicted detection boxes; t k represents the predicted coordinates of the kth prediction box, represents the real coordinates corresponding to the k-th prediction box, L reg is the smoothed L1 loss; β is the balancing factor used to balance the two losses.

[0029] In another aspect of the present invention, a target detection system based on big data empowerment and model flow is proposed, and the target detection system includes the following modules:

[0030] A semantic information vector extraction module is configured to obtain the target detection image of the current scene and convert the category information of the image into a semantic information vector through a Word2vector model;

[0031] A matching module configured to perform cosine similarity matching between the semantic information vector and the semantic information vectors of image category information of all public object detection datasets;

[0032] An initialization module, configured to initialize the object detection model of the current scene with the classification fully connected weights of the category with the highest matching value in the super network with variable width and depth;

[0033] The sub-model screening module is configured to pre-specify the number of floating-point operations and the number of parameters of the model, and traverses the model sampling space one by one to obtain K target detection sub-models of the current scene;

[0034] A model training module is configured to pre-train the K object detection sub-models using the training images of the current scene respectively, and use the pre-trained sub-model with the largest mAP value as the final object detection model of the current scene;

[0035] The target detection module is configured to perform target detection on the target detection image of the current scene through the final target detection model of the current scene to obtain a target detection result.

[0036] According to a third aspect of the present invention, an electronic device is provided, comprising:

[0037] at least one processor; and

[0038] a memory communicatively connected to at least one of the processors; wherein,

[0039] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned target detection method based on big data empowerment and model flow.

[0040] In a fourth aspect of the present invention, a computer-readable storage medium is proposed, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to be executed by the computer to implement the above-mentioned target detection method based on big data empowerment and model flow.

[0041] Beneficial effects of the present invention:

[0042] (1) The target detection method of the present invention is based on big data empowerment and model flow. By integrating publicly available large-scale target detection data sets, a supernet that can dynamically change from depth to width is constructed. By empowering the supernet with big data, the adaptability of the model can be greatly improved, and the effect of training hundreds of thousands of sub-models at the same time can be achieved.

[0043] (2) The target detection method of the present invention is based on big data empowerment and model flow. The trained model has extremely strong generalization ability because it has been trained with a data set of millions of levels. It can be used as an excellent initialization model for actual downstream deployment scenarios. The specific deployment scenario only needs to provide a small amount of labeled data, and the model can achieve very excellent performance.

[0044] (3) The target detection method of the present invention is based on big data empowerment and model flow. It utilizes the fact that images with the same domain distribution have similar feature maps in the fully connected layer of the network. It matches the images provided by the downstream deployment scenario from the database trained by the supernet, selects the closest annotated images and adds them to the training of the model deployment, thereby further improving the performance of the model.

[0045] (4) The target detection method of the present invention is based on big data empowerment and model flow. In the initial supernet training process, through the design of dynamic structure, it realizes the simultaneous training of subnets. When deployed in various scenarios, it only needs to directly extract from the trained subnets according to the various requirements of the corresponding scenarios, saving the model deployment consumption in various environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0047] Figure 1 It is a flow chart of a target detection method based on big data empowerment and model flow of the present invention;

[0048] Figure 2 It is a schematic diagram of the framework of the target detection method based on big data empowerment and model flow of the present invention;

[0049] Figure 3 It is a schematic diagram of the search space when a residual neural network is used as a backbone network in an embodiment of a target detection method based on big data empowerment and model flow of the present invention;

[0050] Figure 4 It is a schematic diagram of similar image extraction in a sparse data scenario of an embodiment of a target detection method based on big data empowerment and model flow of the present invention. DETAILED DESCRIPTION

[0051] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.

[0052] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0053] The present invention provides a target detection method based on big data empowerment and model flow. Aiming at the difficulties encountered by the prior art in target detection problems, on the one hand, the current target detection model has relatively low performance in a few sample scenarios, and on the other hand, the complexity of model deployment and waste of resources are addressed. By integrating a large target detection data set and building a flexible dynamic supernet, hundreds of thousands of sub-model training can be completed at one time. In a specific usage scenario, the weights corresponding to the closest semantic categories in the training database are selected according to the word vector model for initialization, and excellent target detection performance can be achieved by fast fine-tuning using a small amount of labeled data in the usage scenario. In the case where the labeled data for the corresponding usage scenario is relatively sparse, similar data will be selected from the training database based on the distribution characteristics of the labeled data in the feature map of the fully connected layer of the network to assist in deployment training and further enhance the performance of the model.

[0054] The present invention provides a target detection method based on big data empowerment and model flow, the target detection method comprising:

[0055] Step S10, obtaining the target detection image of the current scene, and converting the category information of the image into a semantic information vector through the Word2vector model;

[0056] Step S20, performing cosine similarity matching between the semantic information vector and the semantic information vectors of image category information of all public object detection datasets;

[0057] Step S30, initializing the object detection model of the current scene using the classification fully connected weights of the category with the highest matching value in the supernet with variable width and depth;

[0058] Step S40, pre-specify the number of floating-point operations and the amount of parameters of the model, traverse the model sampling space one by one, and obtain K target detection sub-models of the current scene;

[0059] Step S50, pre-training the K object detection sub-models using the training images of the current scene respectively, and taking the pre-trained sub-model with the largest mAP value as the final object detection model for the current scene;

[0060] Step S60, performing target detection on the target detection image of the current scene using the final target detection model of the current scene to obtain a target detection result.

[0061] In order to more clearly illustrate the target detection method based on big data empowerment and model flow of the present invention, the following is combined with Figure 1 and Figure 2 Each step in the embodiment of the present invention is described in detail.

[0062] The target detection method based on big data empowerment and model flow in the first embodiment of the present invention, each step is described in detail as follows:

[0063] Step A10, normalize the images in all the public object detection datasets obtained, and convert the category information corresponding to the images into semantic information vectors through the Word2vector model. In one embodiment of the present invention, the short side lengths of all images are uniformly adjusted to 800 pixels.

[0064] Step A20, calculating the cosine similarity between the semantic information vectors of each category, merging the categories whose cosine similarity is greater than a preset threshold, and performing label remapping. In one embodiment of the present invention, the preset threshold is 0.8.

[0065] Step A30, select any deep learning target detection model, and set the width and depth sampling space of the model for the feature extraction backbone network of the model to obtain the model sampling space.

[0066] In one embodiment of the present invention, a two-stage target detector FasterRCNN model is selected as the target detection model, a residual network Resnet is selected as the backbone network of the model, and a Resnet50 network or a Resnet77 network or a Resnet101 network is selected as the model anchor.

[0067] like Figure 3 As shown, it is a schematic diagram of the search space when a residual neural network is used as the backbone network in an embodiment of the target detection method based on big data empowerment and model flow of the present invention, AR50, AR77 and AR101 represent Resnet50, Resnet77 and Resnet101 networks respectively, Dmin represents the depth of the minimum network, Danchor represents the depth of the corresponding anchor model, Dmax represents the depth of the maximum network, Dstep represents the step size of the network that can be dynamically changed from the minimum depth to the maximum depth, Wmin represents the width of the maximum network, Wanchor represents the width of the corresponding anchor model, Wmax represents the depth of the corresponding maximum model, Wstep represents the step size of the network that can be dynamically changed from the minimum width to the maximum width, Smin represents the minimum input resolution of the network, Sanchor represents the reference input resolution, Smax represents the maximum input resolution of the network, and Sstep represents the step size of the network accepting the input resolution from maximum to minimum.

[0068] The label-remapped images are subjected to random multi-scale scaling and multi-angle flipping operations to obtain an expanded image set.

[0069] Step B10, dividing the image set after label remapping (or the expanded image set) into batches of a set size, extracting features of any batch of images through the feature extraction backbone network, and randomly sampling a model in the model sampling space.

[0070] In step B20, the features of the current batch of images are input into the classification and bounding box regression branches of the corresponding random sampling model for forward propagation, and the global loss of the model is calculated, as shown in formula (1):

[0071] L all =λL rcnn +L rpn (1)

[0072] Among them, L rcnn represents the classification and bounding box regression loss of the model, L rpn The region proposal network part of the model has a loss, and λ is a balancing factor used to balance the two losses.

[0073] The classification and bounding box regression loss of the model is expressed as shown in formula (2):

[0074] Among them, k represents the sequence number of the prediction box, p ki represents the predicted probability that the k-th prediction box is predicted as the i-th category, represents the actual probability that the k-th prediction box corresponds to the i-th class label, L cls is the cross entropy loss, N cls Represents the total number of categories included in the data set; t ki represents the predicted coordinates of the kth prediction box, represents the real coordinates corresponding to the k-th prediction box, L reg To smooth the L1 loss, N reg represents the total number of prediction boxes; γ is the balance factor used to balance the two losses.

[0075] The region proposal network part of the model is lost, which is expressed as shown in formula (3):

[0076]

[0077] Among them, k represents the sequence number of the prediction box, p i Represents the predicted probability of whether the k-th prediction box contains an object, represents the actual probability of whether the k-th prediction box contains an object, L cls is the binary cross entropy loss, K represents the number of all predicted detection boxes; t k represents the predicted coordinates of the kth prediction box, represents the real coordinates corresponding to the k-th prediction box, Lreg is the smoothed L1 loss; β is the balancing factor used to balance the two losses.

[0078] Step B30, reducing the global loss by back propagation method and stochastic gradient descent method to update model parameters, iterating training until the set training end condition is reached, and obtaining a super network with variable width and depth.

[0079] Step S10, obtaining the target detection image of the current scene (i.e., the downstream target detection scene to be deployed), and converting the category information of the image into a semantic information vector through the Word2vector model.

[0080] Step S20: performing cosine similarity matching between the semantic information vector and the semantic information vectors of image category information of all public object detection datasets.

[0081] Step S30 , initializing the object detection model of the current scene using the classification fully connected weights of the category with the highest matching value in the supernet with variable width and depth.

[0082] In step S40, the number of floating-point operations (FLOPs, which can be used to measure the complexity of the algorithm / model) and the number of parameters of the model are pre-specified, and the sub-models that meet the restriction requirements are screened out one by one in the model sampling space, and K sub-models are randomly sampled from the screened sub-models to obtain K target detection sub-models for the current scene.

[0083] Step S50, respectively use the training images of the current scene (that is, the data set of the downstream target detection scene to be deployed) to pre-train the K target detection sub-models to obtain mAP, and select the pre-trained sub-model with the best result among the K target detection sub-models according to the mAP result as the final target detection model of the current scene (that is, the final deployment model of the downstream target detection scene to be deployed).

[0084] Step S60, performing target detection on the target detection image of the current scene using the final target detection model of the current scene to obtain a target detection result.

[0085] like Figure 4 As shown, this is a schematic diagram of similar image extraction in a sparse data scenario in an embodiment of the target detection method based on big data empowerment and model flow of the present invention. By providing labeled pictures that are very similar to the deployment scenario from the huge database maintained locally, the performance of the deep learning algorithm will be greatly improved as the amount of data increases.

[0086] The present invention taps into the potential of large-scale data sets, making the trained model have excellent migration capabilities. In other unseen data domains, only a small amount of relevant domain data is needed to achieve excellent target detection performance. In actual application scenarios, there are often sparse data. The present invention not only provides good model initialization, but also uses the characteristics of the distribution of feature maps of images in the network to select the most similar relevant data for the specific data domain deployed from the annotation database used for training, assisting in the deployment and training of specific downstream models, and achieving excellent target detection performance in scenarios with sparse data. .

[0087] Although the various steps in the above embodiment are described in the above-mentioned order, those skilled in the art can understand that in order to achieve the effect of this embodiment, different steps do not have to be executed in such an order. They can be executed simultaneously (in parallel) or in a reverse order. These simple changes are within the scope of protection of the present invention.

[0088] The target detection system based on big data empowerment and model flow according to the second embodiment of the present invention includes the following modules:

[0089] A semantic information vector extraction module is configured to obtain the target detection image of the current scene and convert the category information of the image into a semantic information vector through a Word2vector model;

[0090] A matching module configured to perform cosine similarity matching between the semantic information vector and the semantic information vectors of image category information of all public object detection datasets;

[0091] An initialization module, configured to initialize the object detection model of the current scene with the classification fully connected weights of the category with the highest matching value in the super network with variable width and depth;

[0092] The sub-model screening module is configured to pre-specify the number of floating-point operations and the number of parameters of the model, and traverses the model sampling space one by one to obtain K target detection sub-models of the current scene;

[0093] A model training module is configured to pre-train the K object detection sub-models using the training images of the current scene respectively, and use the pre-trained sub-model with the largest mAP value as the final object detection model of the current scene;

[0094] The target detection module is configured to perform target detection on the target detection image of the current scene through the final target detection model of the current scene to obtain a target detection result.

[0095] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0096] It should be noted that the target detection system based on big data empowerment and model flow provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be decomposed or combined. For example, the modules in the above embodiments can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps and are not regarded as improper limitations of the present invention.

[0097] An electronic device according to a third embodiment of the present invention includes:

[0098] at least one processor; and

[0099] a memory communicatively connected to at least one of the processors; wherein,

[0100] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned target detection method based on big data empowerment and model flow.

[0101] A computer-readable storage medium according to a fourth embodiment of the present invention stores computer instructions, and the computer instructions are used to be executed by the computer to implement the above-mentioned target detection method based on big data empowerment and model flow.

[0102] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the storage device and processing device described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0103] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0104] The terms "first", "second", etc. are used to distinguish similar objects rather than to describe or indicate a particular order or sequence.

[0105] The term "comprise" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that includes a list of elements includes not only those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, article, or apparatus / device.

[0106] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A target detection method based on big data empowerment and model flow, characterized in that: The target detection method includes: Step S10, obtaining the target detection image of the current scene, and converting the category information of the image into a semantic information vector through the Word2vector model; Step S20, performing cosine similarity matching between the semantic information vector and the semantic information vectors of image category information of all public object detection datasets; Step S30, initializing the object detection model of the current scene using the classification fully connected weights of the category with the highest matching value in the supernet with variable width and depth; Step S40, pre-specify the number of floating-point operations and the amount of parameters of the model, traverse the model sampling space one by one, and obtain K target detection sub-models of the current scene; Step S50, pre-training the K target detection sub-models using the training images of the current scene respectively, and taking the pre-trained sub-model with the largest mean Average Precision value as the final target detection model for the current scene; Step S60, performing target detection on the target detection image of the current scene using the final target detection model of the current scene to obtain a target detection result.

2. The target detection method based on big data empowerment and model flow according to claim 1 is characterized in that: The model sampling space is obtained by: Step A10, normalizing the images in all the public object detection datasets obtained, and converting the category information corresponding to the images into semantic information vectors through the Word2vector model; Step A20, calculating the cosine similarity between the semantic information vectors of each category, and merging the categories whose cosine similarity is greater than a preset threshold, and performing label remapping; Step A30, select any deep learning target detection model, and set the width and depth sampling space of the model for the feature extraction backbone network of the model to obtain the model sampling space.

3. The target detection method based on big data empowerment and model flow according to claim 2 is characterized in that: The construction and training method of the variable width and depth supernet is as follows: Step B10, dividing the image set after label remapping into batches of a set size, extracting features of any batch of images through the feature extraction backbone network, and randomly sampling a model in the model sampling space; Step B20, input the features of the current batch of images into the classification and bounding box regression branches of the corresponding random sampling model for forward propagation, and calculate the global loss of the model; Step B30, reducing the global loss by back propagation method and stochastic gradient descent method to update model parameters, iterating training until the set training end condition is reached, and obtaining a super network with variable width and depth.

4. The target detection method based on big data empowerment and model flow according to claim 3 is characterized in that: Before dividing the image set after label remapping into batches of a set size in step B10, an image set expansion step is also provided, and the method is as follows: The label-remapped images are subjected to random multi-scale scaling and multi-angle flipping operations to obtain an expanded image set.

5. The target detection method based on big data empowerment and model flow according to claim 3 is characterized in that: The global loss is expressed as: THE all =λL rcnn +L rpn Among them, L rcnn represents the classification and bounding box regression loss of the model, L rpn The region proposal network part of the model has a loss, and λ is a balancing factor used to balance the two losses.

6. The target detection method based on big data empowerment and model flow according to claim 5 is characterized in that: The classification and bounding box regression losses of the model are expressed as: Among them, k represents the sequence number of the prediction box, p ki represents the predicted probability that the k-th prediction box is predicted as the i-th category, represents the actual probability that the k-th prediction box corresponds to the i-th class label, L cls is the cross entropy loss, N cls Represents the total number of categories included in the data set; t ki represents the predicted coordinates of the kth prediction box, represents the real coordinates corresponding to the k-th prediction box, L reg To smooth the L1 loss, N reg represents the total number of prediction boxes; γ is the balance factor used to balance the two losses.

7. The target detection method based on big data empowerment and model flow according to claim 5 is characterized in that: The region proposal network part of the model has a loss expressed as: Among them, k represents the sequence number of the prediction box, p i Represents the predicted probability of whether the k-th prediction box contains an object, represents the actual probability of whether the k-th prediction box contains an object, L cls is the binary cross entropy loss, K represents the number of all predicted detection boxes; t k represents the predicted coordinates of the kth prediction box, represents the real coordinates corresponding to the k-th prediction box, L reg is the smoothed L1 loss; β is the balancing factor used to balance the two losses.

8. A target detection system based on big data empowerment and model flow, characterized in that: The target detection system includes the following modules: A semantic information vector extraction module is configured to obtain the target detection image of the current scene and convert the category information of the image into a semantic information vector through a Word2vector model; A matching module configured to perform cosine similarity matching between the semantic information vector and the semantic information vectors of image category information of all public object detection datasets; An initialization module, configured to initialize the object detection model of the current scene with the classification fully connected weights of the category with the highest matching value in the super network with variable width and depth; The sub-model screening module is configured to pre-specify the number of floating-point operations and the number of parameters of the model, and traverses the model sampling space one by one to obtain K target detection sub-models of the current scene; A model training module is configured to pre-train the K object detection sub-models using the training images of the current scene respectively, and use the pre-trained sub-model with the largest mAP value as the final object detection model of the current scene; The target detection module is configured to perform target detection on the target detection image of the current scene through the final target detection model of the current scene to obtain a target detection result.

9. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the target detection method based on big data empowerment and model flow as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to be executed by the computer to implement the target detection method based on big data empowerment and model flow as described in any one of claims 1-7.

Images