An identification and detection method, device and system for target partial occlusion

Abstract

The present invention discloses a recognition and detection method, device and system for a target with partial occlusion. The device includes an image acquisition unit and a filling recognition unit. The system includes a recognition and detection module, a data storage module and a user interaction module. By judging the first occlusion area ratio of the first part of the occluded part, when the first occlusion area ratio is less than the first threshold, fitting and filling the first part, and performing whole-image recognition on the first processed image obtained by filling, the recognition and detection method, device and system improve the recognition accuracy of the target with partial occlusion; further, a recognition and detection method, device and system for a target with partial occlusion provided by the present invention also, when the first occlusion area ratio is less than the first threshold, perform tile recognition on the captured image through a pre-trained tile recognition model to obtain a recognition result, thereby further improving the recognition accuracy when the target occlusion area is large.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and relates to a recognition and detection method, device and system for a target with partial occlusion. Background Art

[0002] Image processing technology refers to the technology of analyzing the image to be processed by a computer to achieve the required purpose. Among them, the image to be processed usually refers to a digital image, which is a large two-dimensional array obtained by shooting with devices such as industrial cameras, video cameras, and scanners. The elements of the array are called pixels, and their values are called gray values. Image processing technology generally includes image compression, enhancement, restoration, image matching, description and recognition. Among them, due to the development of digitization and intelligence, image recognition is becoming more and more important in the field of image processing.

[0003] In the prior art, image recognition is usually achieved through the steps of image acquisition, image preprocessing, feature extraction, and image recognition in sequence. Among them, image preprocessing refers to the operation performed on the image to be recognized in order to eliminate irrelevant information in the image and restore useful real information, so as to improve the reliability of subsequent feature extraction; and image recognition is the process of comparing the features extracted in the feature extraction step with the stored features, and then taking the feature image with the highest similarity as the recognition result.

[0004] However, the prior art still has the following defects: when there is an occluded part in the target, the recognition accuracy will decrease significantly, and it is difficult to accurately recognize the target.

[0005] Therefore, there is a need for a recognition and detection method, device and system for a target with partial occlusion to solve the above problems existing in the prior art. Summary of the Invention

[0006] Aiming at the existing above technical problems, the purpose of the present invention is to provide a recognition and detection method, device and system for a target with partial occlusion, so as to improve the recognition accuracy of the target with partial occlusion.

[0007] The present invention provides a recognition and detection method for a target with partial occlusion, including: obtaining a captured image of the target to be recognized, determining a first occlusion area ratio of a first part of the target to be recognized that is occluded, and determining whether the first occlusion area ratio is less than a preset first threshold; when the first occlusion area ratio is less than the first threshold, performing fitting filling on the captured image according to a preset target feature library to be recognized and a preset fitting recognition method to obtain a first processed image; and performing recognition on the first processed image according to a preset whole-image recognition method and a preset whole-image recognition model to obtain a recognition result.

[0008] In one embodiment, the recognition and detection method further includes: when the first occlusion area ratio is not less than the first threshold, obtaining a first coordinate range of the first part in the target to be recognized; and performing tile recognition on the captured image according to the first coordinate range and a preset tile recognition model to obtain a recognition result.

[0009] In one embodiment, the recognition and detection method further includes: collecting a plurality of first original images of the target to be recognized from multiple channels, and training a preset neural network classification model with the plurality of first original images according to a preset first model training method to obtain a whole-image recognition model; dividing each of the first original images into several blocks to obtain a plurality of block data sets, and training the preset neural network classification model with the plurality of block data sets according to a preset second model training method to obtain a tile recognition model.

[0010] In one embodiment, when the first occlusion area ratio is less than the first threshold, performing fitting filling on the captured image according to a preset target-to-be-recognized feature library and a preset fitting recognition method to obtain a first processed image, specifically: obtaining a second coordinate range of the first part in the target to be recognized, and retrieving a second feature group with the highest correlation degree with the second coordinate range from the preset target-to-be-recognized feature library; fitting a first fitting image of the first part according to the second feature group and a preset fitting method, and filling the first fitting image into a first area corresponding to the first part in the captured image to obtain a first fitting image; and performing smoothing processing on an edge of the first area in the first fitting image to obtain a first processed image.

[0011] The present invention also provides a recognition and detection device for target partial occlusion. The recognition and detection device includes an image acquisition unit and a filling and recognition unit. The image acquisition unit is configured to acquire a captured image of the target to be recognized, determine a first occlusion area ratio of a first occluded part of the target to be recognized, and determine whether the first occlusion area ratio is less than a preset first threshold. The filling and recognition unit is configured to, when the first occlusion area ratio is less than the first threshold, perform fitting filling on the captured image according to a preset target-to-be-recognized feature library and a preset fitting recognition method to obtain a first processed image; and perform recognition on the first processed image according to a preset whole-image recognition method and a preset whole-image recognition model to obtain a recognition result.

[0012] In one embodiment, the recognition and detection device further includes a tile recognition unit, and the tile recognition unit is configured to: when the first occlusion area ratio is not less than the first threshold, obtain a first coordinate range of the first part in the target to be recognized; according to the first coordinate range and a preset tile recognition method, perform tile recognition on the captured image to obtain a recognition result.

[0013] In one embodiment, the recognition and detection device further includes a model training unit, and the model training unit is configured to: collect a plurality of first original images of the target to be recognized from multiple channels, and according to a preset first model training method, train a preset neural network classification model through the plurality of first original images to obtain a first whole-image recognition model; divide each first original image into several blocks to obtain a plurality of block data sets, and according to a preset second model training method, train the preset neural network classification model through the plurality of block data sets to obtain a first whole-image recognition model.

[0014] The present invention also provides a recognition and detection system for a partially occluded target. The recognition and detection system includes a recognition and detection module, a data storage module, and a user interaction module. The recognition and detection module, the data storage module, and the user interaction module are communicatively connected to each other. Among them, the recognition and detection module is configured to execute the recognition and detection method for a partially occluded target as described above to obtain a recognition result; the data storage module is configured to store a target to be recognized feature library, a whole-image recognition model, and a tile recognition model; the user interaction module is configured to send the recognition result to the user.

[0015] In one embodiment, the recognition and detection module includes a camera, and the user interaction module includes a communication module; the camera is configured to obtain a captured image of the target to be recognized, and the communication module is configured to send the recognition result to the user.

[0016] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0017] The present invention provides a recognition and detection method, device, and system for a partially occluded target. By judging the first occlusion area ratio of the first part of the occluded part, when the first occlusion area ratio is less than the first threshold, fitting and filling the first part, and performing whole-image recognition on the first processed image obtained by filling, the recognition and detection method, device, and system improve the recognition accuracy of the partially occluded target.

[0018] Furthermore, when the first occlusion area ratio is less than the first threshold, the recognition detection method, device and system for target partial occlusion provided by the present invention further perform tile recognition on the captured image through a pre-trained tile recognition model to obtain a recognition result, thereby further improving the recognition accuracy when the target occlusion area is large. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention will be further described below in conjunction with the accompanying drawings of the specification, where:

[0020] Figure 1 shows a flowchart of an embodiment of a recognition detection method for target partial occlusion according to the present invention;

[0021] Figure 2 shows a flowchart of another embodiment of a recognition detection method for target partial occlusion according to the present invention;

[0022] Figure 3 shows a structural diagram of an embodiment of a recognition detection device for target partial occlusion according to the present invention;

[0023] Figure 4 shows a structural diagram of an embodiment of a recognition detection system for target partial occlusion according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention. Specific Embodiment 1

[0026] The embodiment of the present invention first describes a recognition detection method for target partial occlusion. Figure 1 shows a flowchart of an embodiment of a recognition detection method for target partial occlusion according to the present invention.

[0027] As Figure 1 shown, the method includes the following steps:

[0028] S1: Obtain a captured image of the target to be recognized, determine the first occlusion area ratio of the first occluded part of the target to be recognized, and judge whether the first occlusion area ratio is less than a preset first threshold.

[0029] When the target to be recognized is partially occluded, the recognition methods in the prior art often have difficulty in recognition or have low recognition accuracy. In this regard, the embodiments of the present invention describe a recognition method when the target to be recognized is partially occluded to improve the accuracy. This method first classifies the captured image including the target to be recognized according to the area ratio of occlusion, and adopts different methods for different categories to identify, so as to improve the accuracy.

[0030] Specifically, the embodiments of the present invention first obtain the captured image of the target to be recognized, so as to determine the first occlusion area ratio between the area of the first occluded part and the area of the entire target to be recognized, and judge whether the first occlusion area ratio is less than a preset first threshold. When it is less than the preset first threshold, it means that in this captured image, the degree of occlusion of the target to be recognized is not high, and the picture can be appropriately fitted and filled, and then the whole picture is recognized; when the first occlusion area ratio is not less than the preset first threshold, it means that the target to be recognized has been highly occluded, and at this time, the accuracy of using the whole picture recognition will be greatly reduced.

[0031] S2: When the first occlusion area ratio is less than the first threshold, according to the preset feature library of the target to be recognized and the preset fitting recognition method, the captured image is fitted and filled to obtain a first processed image.

[0032] Specifically, when it is judged according to the first occlusion area ratio that the degree of occlusion of the target to be recognized is not high, the embodiments of the present invention perform a fitting and filling process on the captured image for subsequent whole picture recognition. First, obtain the occlusion range of the first part, then, retrieve the feature information associated with the occlusion range of the first part from the existing feature database of the target to be recognized, and then, by means of fitting and filling, make up for the first occluded part of the target to be recognized in the captured image, so that the recognizable area of the target to be recognized in the captured image is complete. Finally, use the preset noise processing method to perform denoising and smoothing processing on the splicing part between the fitted regional image and the first part in the captured image.

[0033] Specifically, in one embodiment, when the first occlusion area ratio is less than the first threshold, according to a preset target feature library to be recognized and a preset fitting recognition method, the captured image is fitted and filled to obtain a first processed image, which is specifically as follows: Obtain the second coordinate range of the first part in the target to be recognized, and retrieve the second feature group with the highest correlation degree with the second coordinate range in the preset target feature library to be recognized; According to the second feature group and the preset fitting method, fit to obtain a first fitting image of the first part, and fill the first fitting image into the first area corresponding to the first part in the captured image, so as to obtain a first fitting image; Smooth the edge of the first area in the first fitting image, so as to obtain a first processed image.

[0034] S3: According to a preset whole-image recognition method and a preset whole-image recognition model, recognize the first processed image to obtain a recognition result.

[0035] After obtaining the first processed image, the first processed image can be recognized as a whole through a pre-trained whole-image recognition model to obtain a recognition result. Among them, since in step S2, the existing target feature map to be recognized of the target to be recognized is used for fitting and filling, therefore, the requirement for the recognition accuracy of the recognition result output during the training process of the whole-image recognition model should be improved.

[0036] The present invention provides a recognition and detection method for target partial occlusion. By judging the first occlusion area ratio of the first part of the occluded part, when the first occlusion area ratio is less than the first threshold, the first part is fitted and filled, and the first processed image obtained by filling is recognized as a whole. This recognition and detection method improves the recognition accuracy of the target with partial occlusion. Specific Embodiment Two

[0038] Furthermore, the embodiment of the present invention also describes a recognition and detection method for target partial occlusion. Figure 2 The flowchart of another embodiment of a recognition and detection method for target partial occlusion according to the present invention is shown.

[0039] As Figure 2 shown, the method includes the following steps:

[0040] A1: Collect a plurality of first original images of the target to be recognized from multiple channels, and train a preset neural network classification model through the plurality of first original images according to a preset first model training method to obtain a whole-image recognition model.

[0041] In order to perform whole-image recognition on the target to be recognized without the target being highly occluded, the method described in the embodiments of the present invention needs to pre-train a whole-image recognition model for the target to be recognized.

[0042] Specifically, it is necessary to first collect multiple images of the target to be recognized through various channels as the model data set. Subsequently, according to a preset data set division method, the model data set is divided into a training set and a validation set. Finally, the preset neural network classification model is trained with the training set and verified using the validation set to obtain multiple training models. Finally, the training model with the highest accuracy is selected from the multiple training models as the whole-image recognition model. In one embodiment, the various channels include network collection, shooting collection, hand-drawing collection, and so on. In one embodiment, the preset data set division method is the K-fold cross-validation method or other commonly used data set division methods in the art. In one embodiment, the preset neural network classification model is a convolutional neural network model with multi-scale convolutional kernels, which is used to extract multi-scale features from multiple first original images to improve the accuracy of whole-image recognition.

[0043] In addition, since in the subsequent step A41, the existing feature map of the target to be recognized is used for fitting and filling, the recognition accuracy requirement for the recognition result output by the whole-image recognition model during training should be increased, so as to improve the final recognition accuracy.

[0044] A2: Divide each of the first original images into several blocks to obtain a plurality of block data sets, and according to a preset second model training method, train the preset neural network classification model through the plurality of block data sets to obtain a block recognition model.

[0045] After training to obtain the whole-image recognition model, it is also necessary to train and obtain a block recognition model to provide recognition model support when the target to be recognized is highly occluded.

[0046] Considering that when the target to be recognized is highly occluded, forcing fitting will instead cause the image recognition result to deviate from the actual situation. Therefore, in the case where the target to be recognized is highly occluded, the embodiments of the present invention intend to use block recognition to improve the recognition accuracy.

[0047] In practical applications, it is necessary to divide and sort the pre-collected target to be recognized according to the feature strength to obtain a block feature strength sequence of the target to be recognized, where the block feature strength sequence includes the position range and corresponding feature strength of each block. In one embodiment, the sizes of the respective blocks of the target to be recognized may be the same or different. After division, each of the first original images is divided according to the feature strength sequence, and the blocks in the same position range of the multiple first original images are grouped into a block data set, thereby obtaining a plurality of block data sets.

[0048] After obtaining multiple block data sets, each block data set is used to train a preset neural network classification model, so as to correspondingly obtain multiple tile recognition models.

[0049] Specifically, for a block data set, according to a preset second data set division method, the block data set is divided into a training set and a validation set. The block data set is trained and validated through the training set, so as to obtain a block model group, and the block model with the highest recognition accuracy is selected from the block model group as the tile recognition model. This step is performed for each block data set, so as to correspondingly obtain multiple tile recognition models that correspond one by one to the multiple block data sets.

[0050] A3: Obtain a captured image of the target to be recognized, determine a first occlusion area ratio of a first occluded part of the target to be recognized, and determine whether the first occlusion area ratio is less than a preset first threshold.

[0051] When the target to be recognized is partially occluded, the recognition methods in the prior art often have difficulty in recognition or low recognition accuracy. In this regard, the embodiments of the present invention describe a recognition method when the target to be recognized is partially occluded to improve the accuracy. This method first needs to classify the captured image including the target to be recognized according to the occlusion area ratio, and adopt different methods for different categories to recognize, so as to improve the accuracy.

[0052] Specifically, the embodiments of the present invention first obtain a captured image of the target to be recognized, so as to determine a first occlusion area ratio between the area of the first occluded part and the area of the entire target to be recognized, and determine whether the first occlusion area ratio is less than a preset first threshold. When it is less than the preset first threshold, it means that in this captured image, the degree of occlusion of the target to be recognized is not high, and the picture can be appropriately fitted and filled, and then the whole picture can be recognized; when the first occlusion area ratio is not less than the preset first threshold, it means that the target to be recognized has been highly occluded, and the accuracy of using the whole picture recognition at this time will be greatly reduced.

[0053] A41: When the first occlusion area ratio is less than the first threshold, perform fitting and filling on the captured image according to a preset target feature library to be recognized and a preset fitting recognition method to obtain a first processed image.

[0054] Specifically, when it is determined according to the first occlusion area ratio that the occlusion degree of the target to be recognized is not high, the embodiment of the present invention performs fitting filling processing on the captured image for subsequent full-image recognition. First, obtain the occlusion range of the first part. Then, retrieve the feature information associated with the occlusion range of the first part from the existing feature database of the target to be recognized. Subsequently, by means of fitting filling, make up for the occluded first part of the target to be recognized in the captured image so that the recognizable area of the target to be recognized in the captured image is complete. Finally, use the preset noise processing method to perform denoising and smoothing processing on the splicing part between the fitted area image and the first part in the captured image.

[0055] Specifically, in one embodiment, when the first occlusion area ratio is less than the first threshold, according to the preset feature library of the target to be recognized and the preset fitting recognition method, perform fitting filling on the captured image to obtain a first processed image, specifically: obtain the second coordinate range of the first part in the target to be recognized, and retrieve the second feature group with the highest correlation degree with the second coordinate range from the preset feature library of the target to be recognized; according to the second feature group and the preset fitting method, fit to obtain the first fitting image of the first part, and fill the first fitting image into the first area corresponding to the first part in the captured image, so as to obtain the first fitting image; perform smoothing processing on the edge of the first area in the first fitting image, so as to obtain the first processed image.

[0056] A51: Recognize the first processed image according to the preset full-image recognition method and the preset full-image recognition model to obtain a recognition result.

[0057] After obtaining the first processed image, the first processed image can be recognized by the pre-trained full-image recognition model to obtain a recognition result. Among them, since in step A41, the existing feature map of the target to be recognized is used for fitting filling, therefore, the requirement for the recognition accuracy of the recognition result output during the training process of the full-image recognition model should be improved.

[0058] A42: When the first occlusion area ratio is not less than the first threshold, obtain the first coordinate range of the first part in the target to be recognized.

[0059] When the target to be recognized is highly occluded, it is necessary to perform block recognition on the captured image. First, it is necessary to obtain the first coordinate range of the occluded first part in the target to be recognized, so as to determine the first to-be-recognized block corresponding to the available recognition part.

[0060] A52: Identify tiles in the captured image according to the first coordinate range and a preset tile recognition model to obtain a recognition result.

[0061] After obtaining the first coordinate range and determining the first block to be recognized corresponding to the recognizable part of the captured image according to the first coordinate range, obtain the feature intensity values corresponding to multiple first blocks to be recognized according to the aforementioned feature strength sequence list; subsequently, determine whether the feature intensity value of each first block to be recognized is greater than a preset feature intensity threshold. When there is a first block to be recognized whose feature intensity value is greater than the preset feature intensity threshold, select the first block to be recognized with the highest feature intensity and the corresponding tile recognition model for classification recognition to obtain a model result.

[0062] The present invention provides a recognition and detection method for a target with partial occlusion. By judging the first occlusion area ratio of the first part of the occluded part, when the first occlusion area ratio is less than a first threshold, fit and fill the first part, and perform whole-image recognition on the first processed image obtained by filling, this recognition and detection method improves the recognition accuracy of a target with partial occlusion; further, a recognition and detection method for a target with partial occlusion provided by the present invention also, when the first occlusion area ratio is less than the first threshold, perform tile recognition on the captured image through a pre-trained tile recognition model to obtain a recognition result, thereby further improving the recognition accuracy when the target occlusion area is large. Specific Embodiment III

[0064] In addition to the above method, an embodiment of the present invention also describes a recognition and detection device for a target with partial occlusion. Figure 3 Shows a structural diagram of an embodiment of a recognition and detection device for a target with partial occlusion according to the present invention.

[0065] As Figure 3 shown, the recognition and detection device includes an image acquisition unit 11 and a filling and recognition unit 12.

[0066] The image acquisition unit 11 is used to acquire a captured image of the target to be recognized, determine the first occlusion area ratio of the first part of the target to be recognized that is occluded, and judge whether the first occlusion area ratio is less than a preset first threshold.

[0067] The filling and recognition unit 12 is used to, when the first occlusion area ratio is less than the first threshold, perform fitting and filling on the captured image according to a preset target feature library to be recognized and a preset fitting recognition method to obtain a first processed image; and, perform recognition on the first processed image according to a preset whole-image recognition method and a preset whole-image recognition model to obtain a recognition result.

[0068] When it is necessary to identify a captured image with a target partially occluded, the identification and detection device first obtains the captured image of the target to be identified through the image acquisition unit 11, determines the first occlusion area ratio of the first part of the target to be identified that is occluded, and determines whether the first occlusion area ratio is less than a preset first threshold; then, through the filling identification unit 12, when the first occlusion area ratio is less than the first threshold, the captured image is filled by fitting according to a preset target feature library to be identified and a preset fitting identification method to obtain a first processed image, and the first processed image is identified according to a preset whole-image identification method and a preset whole-image identification model to obtain an identification result.

[0069] In one embodiment, the identification and detection device further includes a tile identification unit, and the tile identification unit is configured to: when the first occlusion area ratio is not less than the first threshold, obtain the first coordinate range of the first part in the target to be identified; according to the first coordinate range and a preset tile identification method, perform tile identification on the captured image to obtain an identification result.

[0070] In one embodiment, the identification and detection device further includes a model training unit, and the model training unit is configured to: collect a plurality of first original images of the target to be identified from multiple channels, and train a preset neural network classification model through the plurality of first original images according to a preset first model training method to obtain a first whole-image identification model; divide each first original image into several blocks to obtain a plurality of block data sets, and train a preset neural network classification model through the plurality of block data sets according to a preset second model training method to obtain a first whole-image identification model.

[0071] The present invention provides an identification and detection device for a target with partial occlusion. By determining the first occlusion area ratio of the first part of the occluded part, when the first occlusion area ratio is less than the first threshold, the first part is filled by fitting, and the whole image of the filled first processed image is identified. The identification and detection device improves the identification accuracy of the target with partial occlusion; further, an identification and detection device for a target with partial occlusion provided by the present invention also performs tile identification on the captured image through a pre-trained tile identification model when the first occlusion area ratio is less than the first threshold to obtain an identification result, thereby further improving the identification accuracy when the target occlusion area is large. Specific Embodiment Four

[0073] In addition to the above methods and devices, the present invention also describes an identification and detection system for a target with partial occlusion. Figure 4 The structure diagram of an embodiment of an identification and detection system for a target with partial occlusion according to the present invention is shown.

[0074] As Figure 4 shown, the recognition and detection system includes a recognition and detection module 1, a data storage module 2, and a user interaction module 3, and the recognition and detection module 1, the data storage module 2, and the user interaction module 3 are communicatively connected to each other.

[0075] Among them, the recognition and detection module 1 is used to execute the recognition and detection method for the target partially occluded as described above to obtain a recognition result; the data storage module 2 is used to store the target feature library to be recognized, the whole-image recognition model, and the patch recognition model; the user interaction module 3 is used to send the recognition result to the user.

[0076] In one embodiment, the recognition and detection module 1 includes a camera, and the user interaction module 3 includes a communication module; the camera is used to acquire a captured image of the target to be recognized, and the communication module is used to remotely send the recognition result to the user.

[0077] In one embodiment, the user interaction module 3 further includes a touchable / non-touchable display screen, an input keyboard, a virtual keyboard, an indicator light, a buzzer, and a combination of one or more of the foregoing, so that the recognition result can be directly displayed or presented to the user.

[0078] The present invention provides a recognition and detection system for a target partially occluded. By judging the first occlusion area ratio of the first part of the occluded part, when the first occlusion area ratio is less than the first threshold, the first part is fitted and filled, and the whole image of the first processed image obtained by filling is recognized. This recognition and detection system improves the recognition accuracy of the target with partial occlusion; further, a recognition and detection system for a target partially occluded provided by the present invention also, when the first occlusion area ratio is less than the first threshold, performs patch recognition on the captured image through a pre-trained patch recognition model to obtain a recognition result, thereby further improving the recognition accuracy when the target occlusion area is large.

[0079] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only specific embodiments of the present invention and is not used to limit the protection scope of the present invention. In particular, it is pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An identification and detection method for a target partially occluded, characterized in that, Including: Obtain a captured image of the target to be recognized, determine the first occlusion area ratio of the first part of the target to be recognized that is occluded, and determine whether the first occlusion area ratio is less than a preset first threshold; When the first occlusion area ratio is less than the first threshold, perform fitting filling on the captured image according to a preset target feature library to be recognized and a preset fitting recognition method to obtain a first processed image; Perform recognition on the first processed image according to a preset full-image recognition method and a preset full-image recognition model to obtain a recognition result; When the first occlusion area ratio is not less than the first threshold, obtain the first coordinate range of the first part in the target to be recognized; Perform tile recognition on the captured image according to the first coordinate range and a preset tile recognition model to obtain a recognition result; Wherein, the tile recognition model is constructed in the following manner: Collect a plurality of first original images of the target to be recognized from various channels, and train a preset neural network classification model through the plurality of first original images according to a preset first model training method to obtain a full-image recognition model; Divide each first original image into several blocks to obtain a plurality of block data sets, and train a preset neural network classification model through the plurality of block data sets according to a preset second model training method to obtain a tile recognition model.

2. The recognition and detection method for target partial occlusion according to claim 1, wherein When the first occlusion area ratio is less than the first threshold, performing fitting filling on the captured image according to a preset target feature library to be recognized and a preset fitting recognition method to obtain a first processed image, specifically: Obtain the second coordinate range of the first part in the target to be recognized, and retrieve the second feature group with the highest degree of association with the second coordinate range in a preset target feature library to be recognized; Fit the first fitting image of the first part according to the second feature group and a preset fitting method, and fill the first fitting image into the first area corresponding to the first part on the captured image to obtain a first fitting image; Perform smoothing processing on the edge of the first area in the first fitting image to obtain a first processed image.

3. An identification and detection device for a target with partial occlusion, characterized in that, The recognition detection device includes an image acquisition unit, a filling recognition unit, a tile recognition unit, and a model training unit. Among them, the image acquisition unit is used to obtain a captured image of the target to be recognized, determine the first occlusion area ratio of the first part of the target to be recognized that is occluded, and determine whether the first occlusion area ratio is less than a preset first threshold; The filling recognition unit is used to, when the first occlusion area ratio is less than the first threshold, perform fitting filling on the captured image according to a preset target feature library to be recognized and a preset fitting recognition method to obtain a first processed image; and perform recognition on the first processed image according to a preset full-image recognition method and a preset full-image recognition model to obtain a recognition result; The tile recognition unit is configured to: when the first occlusion area ratio is not less than the first threshold, obtain a first coordinate range of the first part in the target to be recognized; perform tile recognition on the captured image according to the first coordinate range and a preset tile recognition method to obtain a recognition result; The model training unit is configured to: collect a plurality of first original images of the target to be recognized from multiple channels, and train a preset neural network classification model through the plurality of first original images according to a preset first model training method to obtain a first whole-image recognition model; divide each first original image into several blocks to obtain a plurality of block data sets, and train a preset neural network classification model through the plurality of block data sets according to a preset second model training method to obtain a first whole-image recognition model.

4. An identification and detection system for target partial occlusion, characterized in that The recognition and detection system includes a recognition and detection module, a data storage module, and a user interaction module. The recognition and detection module, the data storage module, and the user interaction module are communicatively connected to each other. Among them, the recognition and detection module is configured to execute the recognition and detection method for partial occlusion of the target according to any one of claims 1-2 to obtain a recognition result; the data storage module is configured to store a target feature library to be recognized, a whole-image recognition model, and a tile recognition model; the user interaction module is configured to send the recognition result to the user.

5. The recognition and detection system for target partial occlusion according to claim 4, characterized in that The recognition and detection module includes a camera, and the user interaction module includes a communication module; the camera is configured to obtain a captured image of the target to be recognized, and the communication module is configured to send the recognition result to the user.

Images