Object identification method and system based on RGB color cat eye

Abstract

The invention discloses an object identification method and system based on an RGB color cat eye. The method comprises the following steps: acquiring an RGB cat eye image of a to-be-identified object collected by a super-photosensitive cat eye lens; performing bilateral filtering processing on the RGB cat eye image to obtain a preprocessed RGB cat eye image; performing multi-scale feature extraction on the preprocessed RGB cat eye image through an image feature extractor based on a deep neural network model to obtain an RGB cat eye image shallow layer feature map, an RGB cat eye image middle layer feature map and an RGB cat eye image deep layer feature map; fusing the RGB cat eye image shallow layer feature map, the RGB cat eye image middle layer feature map and the RGB cat eye image deep layer feature map to obtain RGB cat eye image semantic fusion features; and based on the RGB cat eye image semantic fusion feature, determining a category label of a to-be-identified object. In this way, the problem of low object recognition precision caused by traditional RGB image quality can be avoided, and the accuracy and robustness of object recognition are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent object recognition, and in particular to an object recognition method based on RGB color cat's eyes and a system thereof. Background Art

[0002] Object recognition is an important task in the field of computer vision, which aims to automatically identify different object categories in images or videos through computer algorithms. With the rapid development of computer vision and deep learning technology, object recognition has been widely used in many application fields, such as intelligent monitoring, autonomous driving, face recognition, etc. However, traditional object recognition schemes are mainly based on RGB images, but RGB images often have some limitations. For example, in low light conditions, the quality of RGB images will be reduced, resulting in a decrease in the accuracy of object recognition.

[0003] Therefore, an optimized object recognition scheme is desired. Summary of the invention

[0004] The embodiment of the present invention provides an object recognition method and system based on RGB color cat's eye, which obtains an RGB cat's eye image of an object to be recognized collected by a super-sensitive cat's eye lens; performs bilateral filtering on the RGB cat's eye image to obtain a pre-processed RGB cat's eye image; performs multi-scale feature extraction on the pre-processed RGB cat's eye image through an image feature extractor based on a deep neural network model to obtain a shallow feature map of the RGB cat's eye image, a middle feature map of the RGB cat's eye image, and a deep feature map of the RGB cat's eye image; fuses the shallow feature map of the RGB cat's eye image, the middle feature map of the RGB cat's eye image, and the deep feature map of the RGB cat's eye image to obtain a semantic fusion feature of the RGB cat's eye image; and, based on the semantic fusion feature of the RGB cat's eye image, determines the category label of the object to be recognized. In this way, the problem of low object recognition accuracy caused by the quality of traditional RGB images can be avoided, and the accuracy and robustness of object recognition can be improved.

[0005] The embodiment of the present invention also provides an object recognition method based on RGB color cat's eye, which includes:

[0006] Obtaining an RGB cat's eye image of the object to be identified collected by a super-sensitive cat's eye lens;

[0007] Performing bilateral filtering on the RGB cat's eye image to obtain a pre-processed RGB cat's eye image;

[0008] Performing multi-scale feature extraction on the pre-processed RGB cat's eye image by an image feature extractor based on a deep neural network model to obtain a shallow feature map of the RGB cat's eye image, a middle feature map of the RGB cat's eye image, and a deep feature map of the RGB cat's eye image;

[0009] Fusing the shallow feature map of the RGB cat's eye image, the middle feature map of the RGB cat's eye image, and the deep feature map of the RGB cat's eye image to obtain a semantic fusion feature of the RGB cat's eye image; and

[0010] Based on the semantic fusion features of the RGB cat's eye image, a category label of the object to be identified is determined.

[0011] The embodiment of the present invention further provides an object recognition system based on RGB color cat's eye, which comprises:

[0012] An RGB cat's eye image acquisition module is used to acquire an RGB cat's eye image of an object to be identified collected by a super-sensitive cat's eye lens;

[0013] A bilateral filtering processing module, used for performing bilateral filtering on the RGB cat's eye image to obtain a pre-processed RGB cat's eye image;

[0014] A multi-scale feature extraction module, used for performing multi-scale feature extraction on the pre-processed RGB cat's eye image through an image feature extractor based on a deep neural network model to obtain a shallow feature map of the RGB cat's eye image, a middle feature map of the RGB cat's eye image, and a deep feature map of the RGB cat's eye image;

[0015] A feature map fusion module, used to fuse the shallow feature map of the RGB cat's eye image, the middle feature map of the RGB cat's eye image and the deep feature map of the RGB cat's eye image to obtain a semantic fusion feature of the RGB cat's eye image; and

[0016] The module for determining the category label of the object to be identified is used to determine the category label of the object to be identified based on the semantic fusion features of the RGB cat's eye image. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0018] Figure 1 The present invention provides a flowchart of an object recognition method based on RGB color cat's eyes in an embodiment of the present invention.

[0019] Figure 2 The figure is a schematic diagram of a system architecture of an object recognition method based on RGB color cat's eye provided in an embodiment of the present invention.

[0020] Figure 3It is a schematic diagram of a palm vein-based recognition method provided in an embodiment of the present invention.

[0021] Figure 4 A schematic diagram of a face recognition method provided in an embodiment of the present invention.

[0022] Figure 5 It is a schematic diagram of a recognition method based on the mixture of face and palm provided in an embodiment of the present invention.

[0023] Figure 6 The block diagram of an object recognition system based on RGB color cat's eye provided in an embodiment of the present invention.

[0024] Figure 7 This is an application scenario diagram of an object recognition method based on RGB color cat's eyes provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0026] Unless otherwise specified, all technical and scientific terms used in the embodiments of the present application have the same meaning as those commonly understood by those skilled in the art of the present application. The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit the scope of this application.

[0027] In the embodiments of the present application, it should be noted that, unless otherwise specified and limited, the term "connection" should be understood in a broad sense. For example, it can be an electrical connection or a connection between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.

[0028] It should be noted that the terms "first\second\third" involved in the embodiments of the present application are only used to distinguish similar objects, and do not represent a specific order for the objects. It is understandable that the specific order or sequence of "first\second\third" can be interchanged where permitted. It should be understood that the objects distinguished by "first\second\third" can be interchanged 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.

[0029] Object recognition is an important task in the field of computer vision, which aims to automatically identify different object categories in images or videos through computer algorithms. With the rapid development of computer vision and deep learning technology, object recognition has been widely used in many application fields and has made remarkable progress.

[0030] Computer vision is the study of how to enable computers to understand and interpret images and videos. Object recognition is a key task in computer vision, which involves matching and classifying objects in input images or videos with predefined categories. Objects can be a variety of entities, including faces, vehicles, animals, objects, etc.

[0031] Deep learning technology plays an important role in object recognition. Deep learning is a machine learning method that learns and represents complex features of data by building a multi-layer neural network model. Convolutional Neural Networks (CNN) is the core model of deep learning in the field of computer vision. It can effectively learn local and global features in images and has achieved remarkable results in object recognition tasks.

[0032] Object recognition plays an important role in many application areas. In the field of intelligent monitoring, object recognition can be used to detect and track abnormal behaviors, identify specific people or objects, etc., to improve the effectiveness of the monitoring system. In the field of autonomous driving, object recognition can help vehicles recognize and understand traffic signs, vehicles and pedestrians on the road, thereby achieving intelligent driving and traffic safety. In the field of face recognition, object recognition can be used to identify and verify faces, and is applied to security authentication, face payment and other fields.

[0033] In traditional object recognition schemes, RGB images are often used as input. RGB images consist of three color channels: red, green, and blue. The combination of these channels can represent different color and brightness information. However, RGB images have some limitations under low-light conditions, which may lead to a decrease in object recognition accuracy.

[0034] In low-light conditions, due to insufficient ambient light, detailed information in the image becomes blurred and noise increases, resulting in a decrease in image quality, which is a challenge for object recognition algorithms because the algorithms usually rely on details and texture information in the image for classification and recognition. In addition, in images under low-light conditions, the contrast between the object and the background is weakened, making it difficult to clearly distinguish the edges and shapes of the object, which affects the accuracy and stability of the object recognition algorithm for the object boundary.

[0035] To overcome these problems, researchers have been looking for ways to improve the performance of object recognition in low-light conditions. A common approach is to use more advanced image enhancement techniques, such as image denoising, enhancement, and contrast enhancement, which can improve image quality by increasing image brightness, adjusting contrast, and enhancing details, thereby improving the accuracy of object recognition. Another approach is to use sensors or cameras with better low-light performance to acquire images. For example, some special sensors or cameras, such as infrared sensors, thermal imaging cameras, or cat-eye lenses that provide super-sensitivity, can obtain higher-quality images in low-light conditions. These sensors can capture information that is different from RGB images, such as infrared radiation or thermal energy, thereby providing richer features for object recognition. In addition, the rapid development of deep learning technology in recent years has also provided new opportunities for the improvement of object recognition in low-light conditions. By using deep learning models such as deep convolutional neural networks (CNNs), high-level features in images can be learned, thereby improving the accuracy of object recognition in low-light conditions.

[0036] In one embodiment of the present invention, Figure 1 The present invention provides a flowchart of an object recognition method based on RGB color cat's eyes in an embodiment of the present invention. Figure 2 Schematic diagram of the system architecture of an object recognition method based on RGB color cat's eye provided in an embodiment of the present invention. Figure 1 and Figure 2 As shown, the object recognition method based on RGB color cat's eye according to an embodiment of the present invention includes: 110, obtaining an RGB cat's eye image of the object to be recognized collected by a super-sensitive cat's eye lens; 120, performing bilateral filtering on the RGB cat's eye image to obtain a pre-processed RGB cat's eye image; 130, performing multi-scale feature extraction on the pre-processed RGB cat's eye image through an image feature extractor based on a deep neural network model to obtain a shallow feature map of the RGB cat's eye image, a middle feature map of the RGB cat's eye image, and a deep feature map of the RGB cat's eye image; 140, fusing the shallow feature map of the RGB cat's eye image, the middle feature map of the RGB cat's eye image, and the deep feature map of the RGB cat's eye image to obtain a semantic fusion feature of the RGB cat's eye image; and, 150, determining the category label of the object to be recognized based on the semantic fusion feature of the RGB cat's eye image.

[0037] Wherein, the deep neural network model is a pyramid network.

[0038] In step 110, an RGB cat's eye image of the object to be identified collected by a super-sensitive cat's eye lens is obtained. Ensure that the image is collected using a super-sensitive cat's eye lens, which can provide clearer and brighter images under low light conditions. In step 120, bilateral filtering is a filtering method that can retain edge information, reduce noise in the image, and maintain the details of the image. The pre-processed RGB cat's eye image can reduce the impact of noise and improve the accuracy of subsequent feature extraction and object recognition. In step 130, a deep neural network-based image feature extractor, such as a convolutional neural network (CNN), is used to extract features from the pre-processed RGB cat's eye image. Multi-scale feature extraction can capture different feature information of the image at different levels. The shallow, middle and deep feature maps of the RGB cat's eye image provide feature representations at different levels, which can better capture the local and global features of the object and enhance the ability to recognize the object. In step 140, the shallow, middle and deep feature maps are fused by an appropriate fusion method (such as weighted average or cascade connection of feature maps) to obtain the semantic fusion features of the RGB cat's eye image. The semantic fusion feature combines feature information at different levels, can provide a more comprehensive and rich feature representation, and helps to improve the accuracy and robustness of object recognition. In step 150, the semantic fusion feature of the RGB cat's eye image is input into a classifier (such as a support vector machine, a random forest or a deep learning model) to perform object recognition and determine the category label of the object to be recognized. Using the semantic fusion feature for object recognition can improve the accuracy of classification because the feature contains richer context and semantic information, which can better distinguish objects of different categories.

[0039] Through the processing of the above steps, we can make full use of the images collected by the super-sensitive cat-eye lens, and improve the accuracy and robustness of object recognition through feature extraction and fusion methods, so as to obtain more reliable category labels.

[0040] In view of the above technical problems, in the technical solution of this application, an object recognition method based on RGB color cat's eye is proposed. RGB color cat's eye is a special camera that uses an ultra-sensitive cat's eye lens and can obtain high-quality color images under low-light conditions. Traditional object recognition methods usually use RGB images as input, while RGB color cat's eye provides better image quality and enhancement capabilities under low-light conditions, so it can play an important role in object recognition tasks.

[0041] Based on this, the technical concept of this application is to collect the RGB cat's eye image of the object to be identified through the super-sensitive cat's eye lens, and introduce image processing and analysis algorithms at the back end to perform feature analysis of the RGB cat's eye image, so as to perform object identification and detection. In this way, the problem of low object recognition accuracy caused by the quality of traditional RGB images can be avoided, the accuracy and robustness of object recognition can be improved, and it has broad application prospects.

[0042] Specifically, in the technical solution of the present application, first, an RGB cat's eye image of the object to be identified is acquired by a super-sensitive cat's eye lens. In the process of acquiring the RGB cat's eye image of the object to be identified, various types of noise may exist in the image due to the influence of factors such as lighting conditions and sensor noise, and these noises will interfere with the analysis and processing of the image by the object recognition algorithm. In addition, considering that bilateral filtering is a nonlinear filtering method, it can effectively remove noise while maintaining edge information, making the image clearer. Therefore, in the technical solution of the present application, it is necessary to perform bilateral filtering on the RGB cat's eye image to obtain a pre-processed RGB cat's eye image. It should be understood that bilateral filtering can not only remove noise, but also smooth the image. It takes into account the spatial distance between pixels and the similarity between pixel values, and filters the pixels by weighted averaging. In this way, the subsequent feature extraction and classification algorithms can be made more stable, while the edge information in the image can be retained, making the boundary of the object clearer.

[0043] Then, a convolutional neural network model with excellent performance in extracting implicit features of images is used to perform feature mining of the preprocessed RGB cat's eye image. In particular, in order to obtain feature representations of different levels in the preprocessed RGB cat's eye image to improve the performance and robustness of object recognition, in the technical solution of the present application, the preprocessed RGB cat's eye image is further subjected to an image feature extractor based on a pyramid network to obtain a shallow feature map of the RGB cat's eye image, a middle feature map of the RGB cat's eye image, and a deep feature map of the RGB cat's eye image. It should be understood that the shallow feature map of the RGB cat's eye image contains low-level image features, such as edges, textures, etc. The shallow feature map is useful for capturing local details and texture feature information about the object to be recognized in the image. The middle feature map of the RGB cat's eye image can capture higher-level semantic information, such as shape, structure, etc. The middle feature map has a certain effect on the overall feature representation of the object with a stone tablet. The deep feature map of the RGB cat's eye image has more abstract and advanced semantic information, and can represent more complex features of the object to be recognized. The deep feature map plays an important role in the classification and recognition of the object.

[0044] In the process of image feature extraction based on pyramid networks, in order to improve computational efficiency and reduce memory consumption, the spatial resolution of the feature map is usually gradually reduced through pooling operations, which can reduce the amount of computation while retaining important semantic information. However, reducing the resolution will also lead to the loss of detail information of the feature map. Therefore, when performing object recognition or segmentation tasks, it is necessary to restore the spatial resolution of the feature map to the size of the original image in order to more accurately locate and identify objects. Based on this, in the technical solution of the present application, the RGB cat's eye image deep feature map is further upsampled to obtain an upsampled RGB cat's eye image deep feature map, so as to restore the spatial resolution of the RGB cat's eye image deep feature map to the same size as the RGB cat's eye image middle-layer feature map, so as to perform subsequent feature fusion and object recognition tasks.

[0045] Furthermore, considering that in the process of image feature extraction based on the pyramid network, feature maps at different levels have different image semantic feature information. Therefore, in order to fuse the feature information of different levels about the object to be identified in the RGB cat's eye image to obtain richer and more comprehensive semantic information and improve the performance of object recognition, it is necessary to further fuse the deep feature map of the upsampled RGB cat's eye image and the middle feature map of the RGB cat's eye image to obtain the semantic feature map of the RGB cat's eye image. In particular, the deep feature map of the upsampled RGB cat's eye image has a higher level of semantic information, while the middle feature map of the RGB cat's eye image can provide more local and overall features. By fusing the deep feature map of the upsampled RGB cat's eye image and the middle feature map of the RGB cat's eye image, the semantic information at different levels can be complemented and enhanced with each other, thereby increasing the richness of the image semantic feature information and obtaining a more accurate and robust semantic feature map.

[0046] Next, considering that the shallow feature map of the RGB cat's eye image contains low-level image details and texture information, but its semantic information is relatively weak, therefore, in order to enhance the feature expression ability of the object to be identified in the image for more accurate identification and detection, in the technical solution of the present application, a residual information enhancement fusion module is further used to fuse the RGB cat's eye image semantic feature map and the RGB cat's eye image shallow feature map to obtain a semantic mask enhanced RGB cat's eye image shallow feature map, so as to pass high-level image semantic information to the shallow feature map, thereby enhancing its semantic expression ability. By using the residual information enhancement fusion module, the semantic feature map can be residually connected with the shallow feature map, so that the network can directly learn the difference and incremental information between the two. This method can avoid the loss and blurring of information, help to retain the details and texture information of the original shallow features, and fuse the semantic information into it, thereby providing a more global and more semantic feature representation, and enhancing the robustness of the shallow feature map to factors such as lighting and perspective changes, which helps to improve the performance of the object recognition algorithm in complex scenes.

[0047] In a specific embodiment of the present application, the RGB cat's eye image shallow feature map, the RGB cat's eye image middle feature map and the RGB cat's eye image deep feature map are fused to obtain the RGB cat's eye image semantic fusion feature, including: upsampling the RGB cat's eye image deep feature map to obtain an upsampled RGB cat's eye image deep feature map; fusing the upsampled RGB cat's eye image deep feature map and the RGB cat's eye image middle feature map to obtain the RGB cat's eye image semantic feature map; and performing residual information fusion enhancement processing on the RGB cat's eye image semantic feature map and the RGB cat's eye image shallow feature map to obtain a semantic mask enhanced RGB cat's eye image shallow feature map as the RGB cat's eye image semantic fusion feature.

[0048] Furthermore, the RGB cat's eye image semantic feature map and the RGB cat's eye image shallow feature map are subjected to residual information fusion enhancement processing to obtain a semantic mask enhanced RGB cat's eye image shallow feature map as the RGB cat's eye image semantic fusion feature, including: using a residual information enhancement fusion module to fuse the RGB cat's eye image semantic feature map and the RGB cat's eye image shallow feature map to obtain the semantic mask enhanced RGB cat's eye image shallow feature map.

[0049] Then, the semantic mask enhanced RGB cat's eye image shallow feature map is passed through a classifier to obtain a classification result, and the classification result is used to represent the category label of the object to be identified. Specifically, in the technical solution of the present application, the label of the classifier is the category label of the object to be identified. Therefore, after obtaining the classification result, the identity of the object to be identified can be identified based on the classification result. In this way, the problem of low object recognition accuracy caused by the quality of traditional RGB images can be avoided, thereby improving the accuracy and robustness of object recognition, and has broad application prospects.

[0050] In a specific embodiment of the present application, based on the semantic fusion features of the RGB cat's eye image, determining the category label of the object to be identified includes: passing the semantic mask enhanced RGB cat's eye image shallow feature map through a classifier to obtain a classification result, and the classification result is used to represent the category label of the object to be identified.

[0051] In the technical solution of the present application, after the preprocessed RGB cat's eye image passes through an image feature extractor based on a pyramid network, the shallow feature map of the RGB cat's eye image, the middle feature map of the RGB cat's eye image, and the deep feature map of the RGB cat's eye image obtained express the cross-depth image semantic features of the preprocessed RGB cat's eye image at different scales based on the pyramid network. In this way, after fusing the upsampled RGB cat's eye image deep feature map and the RGB cat's eye image middle feature map to obtain the RGB cat's eye image semantic feature map, and using the residual information enhancement fusion module to fuse the RGB cat's eye image semantic feature map and the RGB cat's eye image shallow feature map, the obtained semantic mask enhanced RGB cat's eye image shallow feature map has not only the multi-scale and multi-depth image semantic fusion feature expression, but also the deep-shallow residual image semantic feature representation based on residual information enhancement, so that the semantic mask enhanced RGB cat's eye image shallow feature map has the cross-scale and cross-depth image semantic feature residual-fusion correlation distribution property in the overall image semantic feature expression dimension, which makes the probability convergence and divergence exist when the semantic mask enhanced RGB cat's eye image shallow feature map is classified and regressed through the classifier, that is, the accuracy of the classification result is reduced.

[0052] Therefore, when the semantic mask enhanced RGB cat's eye image shallow feature map is classified by a classifier, the present application first optimizes the semantic mask enhanced RGB cat's eye image shallow feature map to obtain an optimized semantic mask enhanced RGB cat's eye image shallow feature map, including the following steps:

[0053] Determine the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the feature set of the shallow feature map of the semantic mask enhanced RGB cat's eye image, and divide the median eigenvalue by the difference between the maximum eigenvalue and the minimum eigenvalue to obtain the shallow distribution probability value of the semantic mask enhanced RGB cat's eye image, that is, p=f mid / (f max -f min ), where f mid 、f max 、f min They respectively represent the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the feature set of the shallow feature map of the semantic mask enhanced RGB cat's eye image, and p represents the shallow distribution probability value of the semantic mask enhanced RGB cat's eye image;

[0054] Each eigenvalue in the shallow feature map of the semantic mask enhanced RGB cat's eye image is normalized to its maximum value to obtain the shallow probability eigenvalue f of the semantic mask enhanced RGB cat's eye image i ;

[0055] The shallow probability feature value f of each semantic mask enhanced RGB cat's eye image is calculated by subtracting the shallow distribution probability value p of the semantic mask enhanced RGB cat's eye image as an exponent. i The power function of the first semantic mask is used to enhance the shallow probability convergence eigenvalue f of the RGB cat eye image. 1i =f i 1-p , where p represents the shallow distribution probability value of the semantic mask enhanced RGB cat eye image, and f i represents the shallow probability feature value of the semantic mask enhanced RGB cat eye image, f 1i It represents the shallow probability convergence eigenvalue of the first semantic mask-enhanced RGB cat-eye image;

[0056] The shallow distribution probability value p of the semantic mask enhanced RGB cat's eye image is used as an index to calculate a probability characteristic value f of the shallow layer of each semantic mask enhanced RGB cat's eye image. i The power function of the difference is used to obtain the second semantic mask to enhance the shallow probability convergence eigenvalue f of the RGB cat eye image 2i =(1-f i ) p , where f 2i It indicates that the second semantic mask enhances the shallow probability convergence eigenvalue of the RGB cat's eye image;

[0057] Each of the semantic masks is used to enhance the shallow probability feature value f of the RGB cat eye image i The first semantic mask enhanced RGB cat's eye image shallow probability defined feature value f is multiplied by the difference between one and the shallow distribution probability value p of the semantic mask enhanced RGB cat's eye image to obtain the first semantic mask enhanced RGB cat's eye image shallow probability defined feature value f3i =f i ×(1-p), where f 3i It represents the shallow probability-limited feature value of the first semantic mask to enhance the RGB cat's eye image;

[0058] Combine a and each of the semantic masks to enhance the shallow probability feature value f of the RGB cat eye image i The difference between the values ​​of and is multiplied by the shallow distribution probability value p of the semantic mask enhanced RGB cat's eye image to obtain the shallow probability defined feature value f of the second semantic mask enhanced RGB cat's eye image 4i =(1-f i )×p, where f 4i It represents the second semantic mask to enhance the shallow probability-limited feature value of the RGB cat's eye image;

[0059] After multiplying the shallow probability convergence eigenvalue of the first semantic mask enhanced RGB cat's eye image by the shallow probability convergence eigenvalue of the second semantic mask enhanced RGB cat's eye image, further add the shallow probability limit eigenvalue of the first semantic mask enhanced RGB cat's eye image and the shallow probability limit eigenvalue of the second semantic mask enhanced RGB cat's eye image to obtain the eigenvalue f′ of the shallow feature map of the optimized semantic mask enhanced RGB cat's eye image i =f 1i ×f 2i +f 3i +f 4i , where f′ i Represents the eigenvalues ​​of the shallow feature map of the RGB cat's eye image enhanced by optimizing the semantic mask.

[0060] Finally, the optimized semantic mask-enhanced RGB cat's eye image shallow feature map is passed through a classifier to obtain a classification result.

[0061] Based on this, by using the cross entropy form power series of the regression probability of the shallow feature map of the semantic mask enhanced RGB cat's eye image as the probability distribution convergence limit, the regression probability convergence approximation is performed on the basis of the combination of the feature set distribution and the probability density distribution of the shallow feature map of the semantic mask enhanced RGB cat's eye image, so as to avoid the conservative convergence control of the fatal potential divergence of the shallow feature map of the semantic mask enhanced RGB cat's eye image, and further use the probability distribution cross entropy as the objective function to guide the limited recovery strategy, so as to achieve the common agility of the convergence of the probability convergence and divergence of the shallow feature map of the semantic mask enhanced RGB cat's eye image to the probability density distribution space, thereby improving the accuracy of the classification result.

[0062] In one embodiment of the present application, Figure 3 As shown, a palm vein-based identification method is provided, which includes the steps of:

[0063] S10. The recognition end needs to complete the sensor configuration and adjust the exposure based on this sensor to achieve a clear image of the palm that can display veins and extract features at an appropriate distance.

[0064] S11. First, the palm matrix detection of the palm vein algorithm is used to detect the image and extract the size of the rectangle occupied by the palm in the current image. Secondly, the key information of the rectangular range is extracted, mainly the vein information and a small amount of palm print information. Then, the image quality, including some parameters such as exposure value, palm distance, palm tilt angle, etc., is judged to see whether it meets the algorithm expectations. After that, the palm liveness threshold is evaluated to ensure that it is a real palm.

[0065] S12. After completing the above steps, the key biometric information of the palm is extracted. This information can be saved for comparison with the feature information extracted later. If the comparison score is higher than a certain threshold, it can be considered that the palm is the same person.

[0066] In another embodiment of the present application, Figure 4 As shown, a face recognition method is provided, which includes the steps of:

[0067] S20. Similarly, the sensor configuration needs to be completed for faces. The face exposure adjustment needs to be distinguished from the palm vein exposure. The face can be recognized at a farther distance. By adjusting the appropriate exposure, a clear face image can be seen.

[0068] S21. Based on the clear face image, run the face algorithm library, first detect the image to get the number of faces in the image and the rectangular coordinates of each face. To facilitate the distinction, the face closer to the sensor is used as the basis for recognition. After extracting the key points of the current face, the face quality judgment and liveness judgment are completed.

[0069] S22. Extract key feature value information of the face. This information can be saved and used for comparison with the information extracted each time the face is recognized. If it is above a certain threshold, it is considered to be the same face.

[0070] In another embodiment of the present application, Figure 5 As shown, a recognition method based on the mixture of face and palm is provided, which comprises the steps of:

[0071] S30. The mixed recognition of face and palm depends on the fine processing of image exposure. Based on the face library and palm vein library, the image exposure of the palm is adjusted first. After the palm detection interface is run to complete the palm judgment successfully, the palm recognition step is immediately executed. If there is no palm, the exposure value is adjusted to adapt to the face and the face is detected.

[0072] S31. After completing a round of judgment, if a palm is present, execute the palm recognition step 1 and no longer execute the face recognition step. If a palm is not present, perform face judgment, and if a face is present, execute the face recognition step 2.

[0073] S32. When neither of them is recognized, the palm exposure is changed from the face exposure to the palm exposure to identify the palm. The image exposure is fine-tuned internally. The image exposure during the second round of recognition is different from the previous exposure value. At this time, the operation steps of S30 and S31 are repeated. If neither of them is recognized after a certain period of time, the process stops.

[0074] In summary, the object recognition method based on RGB color cat's eye according to the embodiment of the present invention is explained, which uses a super-sensitive cat's eye lens to collect RGB cat's eye images of the object to be recognized, and introduces image processing and analysis algorithms at the back end to perform feature analysis of the RGB cat's eye image, so as to perform object recognition detection. In this way, the problem of low object recognition accuracy caused by the quality of traditional RGB images can be avoided, the accuracy and robustness of object recognition can be improved, and it has broad application prospects.

[0075] Figure 6 FIG. 1 is a block diagram of an object recognition system based on RGB color cat's eye provided in an embodiment of the present invention. Figure 6 As shown, the object recognition system 200 based on RGB color cat's eye includes: an RGB cat's eye image acquisition module 210, which is used to acquire an RGB cat's eye image of an object to be recognized collected by a super-sensitive cat's eye lens; a bilateral filtering processing module 220, which is used to perform bilateral filtering processing on the RGB cat's eye image to obtain a pre-processed RGB cat's eye image; a multi-scale feature extraction module 230, which is used to perform multi-scale feature extraction on the pre-processed RGB cat's eye image through an image feature extractor based on a deep neural network model to obtain a shallow feature map of the RGB cat's eye image, a middle feature map of the RGB cat's eye image, and a deep feature map of the RGB cat's eye image; a feature map fusion module 240, which is used to fuse the shallow feature map of the RGB cat's eye image, the middle feature map of the RGB cat's eye image, and the deep feature map of the RGB cat's eye image to obtain a semantic fusion feature of the RGB cat's eye image; and a category label determination module 250 for the object to be recognized, which is used to determine the category label of the object to be recognized based on the semantic fusion feature of the RGB cat's eye image.

[0076] In the object recognition system based on RGB color cat's eyes, the deep neural network model is a pyramid network.

[0077] Those skilled in the art will appreciate that the specific operations of each step in the above-mentioned object recognition system based on RGB color cat's eye have been described in detail above. Figures 1 to 5The description of the RGB color cat's eye based object recognition method has been introduced in detail, and therefore, its repeated description will be omitted.

[0078] As described above, the object recognition system 200 based on RGB color cat's eye according to the embodiment of the present invention can be implemented in various terminal devices, such as a server for object recognition based on RGB color cat's eye. In one example, the object recognition system 200 based on RGB color cat's eye according to the embodiment of the present invention can be integrated into the terminal device as a software module and / or a hardware module. For example, the object recognition system 200 based on RGB color cat's eye can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the object recognition system 200 based on RGB color cat's eye can also be one of the many hardware modules of the terminal device.

[0079] Alternatively, in another example, the RGB color cat's eye-based object recognition system 200 and the terminal device may also be separate devices, and the RGB color cat's eye-based object recognition system 200 may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0080] Figure 7 FIG. 1 is an application scenario diagram of an object recognition method based on RGB color cat's eye provided in an embodiment of the present invention. Figure 7 As shown, in this application scenario, first, obtain the super-sensitive cat-eye lens (for example, Figure 7 The RGB cat eye image of the object to be identified (e.g., Figure 7 Then, the obtained RGB cat's eye image is input to a server (for example, Figure 7 In the S) shown in , the server is capable of processing the RGB cat's eye image based on an object recognition algorithm for RGB color cat's eye to determine the category label of the object to be identified.

[0081] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An object recognition method based on RGB color cat's eye, characterized in that: include: Obtaining an RGB cat's eye image of the object to be identified collected by a super-sensitive cat's eye lens; Performing bilateral filtering on the RGB cat's eye image to obtain a pre-processed RGB cat's eye image; Performing multi-scale feature extraction on the pre-processed RGB cat's eye image by an image feature extractor based on a deep neural network model to obtain a shallow feature map of the RGB cat's eye image, a middle feature map of the RGB cat's eye image, and a deep feature map of the RGB cat's eye image; Fusing the shallow feature map of the RGB cat's eye image, the middle feature map of the RGB cat's eye image, and the deep feature map of the RGB cat's eye image to obtain a semantic fusion feature of the RGB cat's eye image; as well as Based on the semantic fusion features of the RGB cat's eye image, a category label of the object to be identified is determined.

2. The object recognition method based on RGB color cat's eye according to claim 1, characterized in that: The deep neural network model is a pyramid network.

3. The object recognition method based on RGB color cat's eye according to claim 2, characterized in that: Fusion of the shallow feature map of the RGB cat's eye image, the middle feature map of the RGB cat's eye image, and the deep feature map of the RGB cat's eye image to obtain a semantic fusion feature of the RGB cat's eye image includes: Upsampling the RGB cat's eye image deep feature map to obtain an upsampled RGB cat's eye image deep feature map; Fusing the upsampled RGB cat's eye image deep feature map and the RGB cat's eye image middle feature map to obtain an RGB cat's eye image semantic feature map; and The RGB cat's eye image semantic feature map and the RGB cat's eye image shallow feature map are subjected to residual information fusion enhancement processing to obtain a semantic mask enhanced RGB cat's eye image shallow feature map as the RGB cat's eye image semantic fusion feature.

4. The object recognition method based on RGB color cat's eye according to claim 3, characterized in that: The residual information fusion enhancement processing is performed on the RGB cat's eye image semantic feature map and the RGB cat's eye image shallow feature map to obtain a semantic mask enhanced RGB cat's eye image shallow feature map as the RGB cat's eye image semantic fusion feature, including: using a residual information enhancement fusion module to fuse the RGB cat's eye image semantic feature map and the RGB cat's eye image shallow feature map to obtain the semantic mask enhanced RGB cat's eye image shallow feature map.

5. The object recognition method based on RGB color cat's eye according to claim 4, characterized in that: Based on the semantic fusion features of the RGB cat's eye image, determining the category label of the object to be identified includes: passing the semantic mask enhanced RGB cat's eye image shallow feature map through a classifier to obtain a classification result, and the classification result is used to represent the category label of the object to be identified.

6. The object recognition method based on RGB color cat's eye according to claim 5, characterized in that: Passing the semantic mask enhanced RGB cat's eye image shallow feature map through a classifier to obtain a classification result, including: Optimizing the semantic mask enhanced RGB cat's eye image shallow feature map to obtain an optimized semantic mask enhanced RGB cat's eye image shallow feature map; The optimized semantic mask-enhanced RGB cat's eye image shallow feature map is passed through a classifier to obtain a classification result.

7. The object recognition method based on RGB color cat's eye according to claim 6, characterized in that: Optimizing the semantic mask enhanced RGB cat's eye image shallow feature map to obtain an optimized semantic mask enhanced RGB cat's eye image shallow feature map, including: Determine the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the feature set of the shallow feature map of the semantic mask enhanced RGB cat's eye image, and divide the median eigenvalue by the difference between the maximum eigenvalue and the minimum eigenvalue to obtain the shallow distribution probability value of the semantic mask enhanced RGB cat's eye image, that is, p=f mid / (f max -f min ), where f mid 、f max 、f min They respectively represent the median eigenvalue, the maximum eigenvalue and the minimum eigenvalue in the feature set of the shallow feature map of the semantic mask enhanced RGB cat's eye image, and p represents the shallow distribution probability value of the semantic mask enhanced RGB cat's eye image; Each eigenvalue in the shallow feature map of the semantic mask enhanced RGB cat's eye image is normalized to its maximum value to obtain the shallow probability eigenvalue f of the semantic mask enhanced RGB cat's eye image i ; The shallow probability feature value f of each semantic mask enhanced RGB cat's eye image is calculated by subtracting the shallow distribution probability value p of the semantic mask enhanced RGB cat's eye image as an exponent. i The power function of the first semantic mask is used to enhance the shallow probability convergence eigenvalue f of the RGB cat eye image. 1i =f i 1-p , where p represents the shallow distribution probability value of the semantic mask enhanced RGB cat eye image, and f i represents the shallow probability feature value of the semantic mask enhanced RGB cat eye image, f 1i It represents the shallow probability convergence eigenvalue of the first semantic mask-enhanced RGB cat-eye image; The shallow distribution probability value p of the semantic mask enhanced RGB cat's eye image is used as an index to calculate a probability characteristic value f of the shallow layer of each semantic mask enhanced RGB cat's eye image. i The power function of the difference is used to obtain the second semantic mask to enhance the shallow probability convergence eigenvalue f of the RGB cat eye image 2i =(1-f i ) p , where f 2i It represents the shallow probability convergence eigenvalue of the second semantic mask-enhanced RGB cat-eye image; Each of the semantic masks is used to enhance the shallow probability feature value f of the RGB cat eye image i The first semantic mask enhanced RGB cat's eye image shallow probability defined feature value f is multiplied by the difference between one and the shallow distribution probability value p of the semantic mask enhanced RGB cat's eye image to obtain the first semantic mask enhanced RGB cat's eye image shallow probability defined feature value f 3i =f i ×(1-p), where f 3i It represents the shallow probability-limited feature value of the first semantic mask to enhance the RGB cat's eye image; Combine a and each of the semantic masks to enhance the shallow probability feature value f of the RGB cat eye image i The difference between the values ​​of and is multiplied by the shallow distribution probability value p of the semantic mask enhanced RGB cat's eye image to obtain the shallow probability defined feature value f of the second semantic mask enhanced RGB cat's eye image 4i =(1-f i )×p, where f 4i It represents the second semantic mask to enhance the shallow probability-limited feature value of the RGB cat's eye image; After multiplying the shallow probability convergence eigenvalue of the first semantic mask enhanced RGB cat's eye image by the shallow probability convergence eigenvalue of the second semantic mask enhanced RGB cat's eye image, further add the shallow probability limit eigenvalue of the first semantic mask enhanced RGB cat's eye image and the shallow probability limit eigenvalue of the second semantic mask enhanced RGB cat's eye image to obtain the eigenvalue f′ of the shallow feature map of the optimized semantic mask enhanced RGB cat's eye image i =f 1i ×f 2i +f 3i +f 4i , where f′ i Represents the eigenvalues ​​of the shallow feature map of the RGB cat's eye image enhanced by optimizing the semantic mask.

8. An object recognition system based on RGB color cat's eye, characterized in that: include: An RGB cat's eye image acquisition module is used to acquire an RGB cat's eye image of an object to be identified collected by a super-sensitive cat's eye lens; A bilateral filtering processing module, used for performing bilateral filtering on the RGB cat's eye image to obtain a pre-processed RGB cat's eye image; A multi-scale feature extraction module, used for performing multi-scale feature extraction on the pre-processed RGB cat's eye image through an image feature extractor based on a deep neural network model to obtain a shallow feature map of the RGB cat's eye image, a middle feature map of the RGB cat's eye image, and a deep feature map of the RGB cat's eye image; A feature map fusion module, used for fusing the shallow feature map of the RGB cat's eye image, the middle feature map of the RGB cat's eye image and the deep feature map of the RGB cat's eye image to obtain a semantic fusion feature of the RGB cat's eye image; as well as The module for determining the category label of the object to be identified is used to determine the category label of the object to be identified based on the semantic fusion features of the RGB cat's eye image.

9. The object recognition system based on RGB color cat's eye according to claim 8, characterized in that: The deep neural network model is a pyramid network.

Images