Intelligent target identification method based on multi-source heterogeneous information fusion

By combining optical and radar to acquire image feature information, the target recognition accuracy problem under the influence of light and occlusion in the prior art is solved, and higher recognition accuracy and robustness are achieved, and target recognition in complex environments is suitable for target recognition .

CN119091268APending Publication Date: 2024-12-06UNIV OF ELECTRONICS SCI & TECH OF CHINA

Patent Information

Application Number
CN202411268171.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-11
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

In the prior art, the acquired images obtained by the image acquisition device are susceptible to light changes and occlusion, resulting in poor accuracy and robustness of the target recognition results.

Method used

An intelligent target recognition method based on multi-source heterogeneous information fusion is adopted to obtain optically collected images and radar-acquisitioned images. By extracting and fusing the feature information of both, a pre-trained target recognition model is used to identify targets. The radar-acquisitioned images are not affected by light and occlusion, and can fully reflect the target environment. The optical-acquisitioned images are clear and colorful, which improves the accuracy and robustness of the recognition results by fusion of feature information.

Benefits of technology

Through the fusion of multi-source heterogeneous information, more complete and true target environment feature information is obtained, which improves the accuracy and robustness of target object information and enhances the target recognition ability in complex environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119091268A_ABST
    Figure CN119091268A_ABST
Patent Text Reader

Abstract

The embodiment of the invention discloses an intelligent target identification method based on multi-source heterogeneous information fusion. The method comprises the following steps: acquiring an optical acquisition image and a radar acquisition image acquired from a target environment; wherein the radar acquisition image is an image generated by the synthetic aperture radar based on a received target echo signal corresponding to the target environment; extracting first feature information corresponding to the optical acquisition image and second feature information corresponding to the radar acquisition image, and performing feature fusion operation on the first feature information and the second feature information to obtain fused feature information; and obtaining target object information of the target object contained in the target environment based on the fusion feature information and a pre-trained target recognition model. According to the technical scheme of the embodiment of the invention, the fusion feature information can fully, completely and truly reflect the features of the target environment; and the accuracy and robustness of the target object information can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of computer technology, and in particular to an intelligent target recognition method based on multi-source heterogeneous information fusion. Background Art

[0002] In recent years, artificial intelligence technologies, represented by intelligent target recognition technology, have been rapidly changing. Intelligent target recognition technology has been widely used in the fields of intelligent security, autonomous driving, etc.

[0003] In the prior art, the captured image acquired by the image acquisition device is usually used as the object to be recognized by the target recognition model, and the captured image is input into the target recognition model to obtain the target recognition result. However, in the process of implementing the present invention, it is found that the prior art has at least the following technical problems: since the captured image acquired by the image acquisition device is easily affected by illumination changes and occlusion, the target recognition result obtained based on the captured image acquired by the image acquisition device has poor accuracy and robustness. Summary of the invention

[0004] The embodiment of the present invention provides an intelligent target recognition method based on multi-source heterogeneous information fusion to achieve the purpose of improving the accuracy and robustness of target object information.

[0005] According to one aspect of the present invention, there is provided an intelligent target recognition method based on multi-source heterogeneous information fusion, comprising:

[0006] Acquire an optical collection image and a radar collection image collected of the target environment; wherein the radar collection image is an image generated by a synthetic aperture radar based on a received target echo signal corresponding to the target environment;

[0007] Extracting first feature information corresponding to the optically collected image and second feature information corresponding to the radar collected image, and performing a feature fusion operation on the first feature information and the second feature information to obtain fused feature information;

[0008] Based on the fused feature information and a pre-trained target recognition model, target object information of the target object contained in the target environment is obtained.

[0009] The technical solution of the embodiment of the present invention is to obtain an optical acquisition image and a radar acquisition image collected from a target environment, wherein the radar acquisition image is an image generated by a synthetic aperture radar based on a received target echo signal corresponding to the target environment; and extract first feature information corresponding to the optical acquisition image and second feature information corresponding to the radar acquisition image, perform a feature fusion operation on the first feature information and the second feature information to obtain fused feature information. Since the radar acquisition image generated by the synthetic aperture radar is not affected by illumination and occlusion, it can more completely and adequately reflect the situation of the target environment, and the optical acquisition image has a clear outline and rich colors, and more realistically reflects the information of the target environment; therefore, by fusing the first feature information with the second feature information, the fused feature information can more fully, completely and realistically reflect the characteristics of the target environment; based on the fused feature information and the pre-trained target recognition model, the target object information of the target object contained in the target environment is obtained, which is conducive to improving the accuracy and robustness of the target object information.

[0010] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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.

[0012] Figure 1 is a flow chart of an intelligent target recognition method based on multi-source heterogeneous information fusion provided according to an embodiment of the present invention;

[0013] Figure 2 is a flow chart of another intelligent target recognition method based on multi-source heterogeneous information fusion provided according to an embodiment of the present invention;

[0014] Figure 3 It is a schematic diagram of the working process of an intelligent target recognition device based on multi-source heterogeneous information fusion provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0016] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "etc." and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0017] It should be noted that the collection, collection, updating, analysis, processing, use, transmission, storage and other aspects of user personal information involved in the technical solution of this disclosure are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information and network security.

[0018] Figure 1 This is a flow chart of an intelligent target recognition method based on multi-source heterogeneous information fusion provided according to an embodiment of the present invention. This embodiment is applicable to the case of identifying a target object in a target environment. The method can be performed by an intelligent target recognition device based on multi-source heterogeneous information fusion, and the intelligent target recognition device based on multi-source heterogeneous information fusion can be implemented in the form of hardware and / or software.

[0019] like Figure 1 As shown, the method of this embodiment may specifically include:

[0020] S110: Acquire optical images and radar images collected from the target environment.

[0021] Among them, the radar acquisition image is an image generated by the synthetic aperture radar based on the received target echo signal corresponding to the target environment. In a specific implementation, a synthetic aperture radar can be used to transmit a target transmission signal in the target environment, and receive a target echo signal corresponding to the target transmission signal, and the image generated based on the target echo signal is used as the radar acquisition image. Since the synthetic radar is not limited by lighting and climatic conditions, it can provide all-day and all-weather observation capabilities. The target environment can be an environment where target object identification is required. For example, in the field of disaster response, the target environment can be an environment where disaster detection is required; in the field of geological exploration, the target environment can be a geological area to be explored. The optical acquisition image can be an image collected by an optical sensor based on optical imaging technology.

[0022] In a specific implementation, an optical acquisition image obtained by an optical sensor after acquiring an image of the target environment may be acquired. In addition, a radar acquisition image generated by a synthetic aperture radar based on a received target echo signal corresponding to the target environment may be acquired.

[0023] S120: extract first feature information corresponding to the optically collected image and second feature information corresponding to the radar collected image, and perform a feature fusion operation on the first feature information and the second feature information to obtain fused feature information.

[0024] The first feature information is used to reflect the image features in the optically collected image, and the second feature information is used to reflect the image features in the radar collected image.

[0025] Specifically, extracting the first feature information corresponding to the optically acquired image and the second feature information corresponding to the radar acquired image includes: extracting the first feature information of the optically acquired image based on a pre-trained first feature extraction model; extracting the second feature information of the radar acquired image based on a pre-trained second feature extraction model; wherein the first feature extraction model and the second feature extraction model include a convolutional neural network model and / or a deep learning model of a self-attention mechanism. The first feature extraction model is a model for extracting features of the optically acquired image, and the second feature extraction model is a model for extracting features of the radar acquired image.

[0026] It should be noted that feature information plays a vital role in image processing. Feature information can help understand the content of the image and achieve the task of target recognition. Traditional feature extraction methods are mostly based on statistical features to extract features such as color, texture and shape. These features have poor performance and cannot accurately extract information of target objects in complex environments or under camouflage. At the same time, the generalization ability is extremely poor, and changing scenes or angles may result in the inability to normally acquire the features of the image. Therefore, this embodiment pre-trains a first feature extraction model and a second feature extraction model, and the first feature extraction model and the second feature extraction model include a convolutional neural network model and / or a deep learning model of a self-attention mechanism. The feature information of optically collected images and radar collected images is extracted through a neural network model, thereby providing a feature extraction method with stronger generalization and expression capabilities.

[0027] Exemplarily, the convolutional neural network can be a residual network, and the deep learning model of the self-attention mechanism includes a Transformer model. For example, when extracting features through a residual network, the residual network forms a convolution module by combining operations such as convolution, padding, and pooling, stacks deep convolutional networks using residual connections, and uses batch normalization to accelerate the network training process, and finally outputs through a fully connected layer structure convolution.

[0028] This embodiment uses a convolutional neural network model and / or a deep learning model of a self-attention mechanism as a feature extraction model to obtain a feature space with stronger generalization and representation capabilities, extract accurate and effective feature information, and facilitate accurate identification of target objects based on the feature information.

[0029] In order to improve the accuracy of the extracted feature information, the optically collected image and the radar collected image may be preprocessed before the feature extraction operation is performed. Specifically, before extracting the first feature information corresponding to the optically collected image, the process further includes: performing at least one image preprocessing operation of image geometric transformation, color space adjustment, noise addition, and image cropping on the optically collected image; and updating the optically collected image based on the image obtained after the image preprocessing operation.

[0030] Optionally, the image geometric transformation includes at least one of image rotation, resizing, horizontal flipping and vertical flipping; the color space adjustment includes at least one of brightness adjustment, contrast adjustment and saturation adjustment; adding noise includes adding Gaussian noise and / or adding salt and pepper noise; and the image cropping includes random cropping and / or center cropping.

[0031] This embodiment provides a specific method corresponding to each image preprocessing operation, thereby facilitating diversified processing of optically collected images.

[0032] In a specific implementation, after performing image preprocessing on the optically captured image, in order to facilitate subsequent feature extraction operations, the image obtained after the image preprocessing operation may be updated to the optically captured image, so as to extract first feature information based on the image preprocessing operation.

[0033] This embodiment improves the image quality and usability of the optically collected image by performing at least one image preprocessing operation of image geometric transformation, color space adjustment, noise addition, and image cropping on the optically collected image.

[0034] Furthermore, before extracting the second feature information corresponding to the radar acquired image, it also includes: performing detail feature enhancement processing on the radar acquired image, and updating the image obtained after the detail feature enhancement processing to the radar acquired image.

[0035] In order to fully and clearly extract the second feature information, detail enhancement processing may be performed on the radar image. Optionally, the detail feature enhancement processing includes at least one of contrast enhancement processing, histogram equalization processing, and transposition processing. In a specific implementation, the image obtained after the detail enhancement processing is updated to the radar image, thereby enhancing the detail features and facilitating the feature extraction operation.

[0036] In addition, before extracting the second feature information corresponding to the radar acquired image, it also includes: based on the spatial adaptive filtering algorithm, performing a filtering preprocessing operation on the radar acquired image, and updating the image obtained after the filtering preprocessing operation to the radar acquired image.

[0037] Among them, the spatial adaptive filtering algorithm is a filtering method widely used in signal processing. It can automatically adjust the filter parameters to achieve the optimal filtering effect when the statistical characteristics of the signal and noise are unknown or change over time. In this embodiment, the spatial adaptive filtering algorithm is used to perform filtering preprocessing operations on radar collected images. Through filtering preprocessing operations, the influence of speckle noise in radar collected images on feature extraction can be reduced and eliminated.

[0038] In this embodiment, in order to simultaneously reflect the feature information of the optically collected image and the radar collected image, feature fusion processing may be performed on the first feature information and the second feature information to obtain fused feature information.

[0039] S130, obtaining target object information of the target object contained in the target environment based on the fused feature information and the pre-trained target recognition model.

[0040] In a specific implementation, the fused feature information can be input into a target recognition model, and the target recognition model outputs target object information of a target object contained in a target environment. The target object information includes the region information and category information to which the target object belongs. The target recognition model includes a convolutional network model.

[0041] The technical solution of the embodiment of the present invention is to obtain an optical acquisition image and a radar acquisition image collected from a target environment, wherein the radar acquisition image is an image generated by a synthetic aperture radar based on a received target echo signal corresponding to the target environment; and extract first feature information corresponding to the optical acquisition image and second feature information corresponding to the radar acquisition image, perform a feature fusion operation on the first feature information and the second feature information to obtain fused feature information. Since the radar acquisition image generated by the synthetic aperture radar is not affected by illumination and occlusion, it can more completely and adequately reflect the situation of the target environment, and the optical acquisition image has a clear outline and rich colors, and more realistically reflects the information of the target environment; therefore, by fusing the first feature information with the second feature information, the fused feature information can more fully, completely and realistically reflect the characteristics of the target environment; based on the fused feature information and the pre-trained target recognition model, the target object information of the target object contained in the target environment is obtained, which is conducive to improving the accuracy and robustness of the target object information.

[0042] Figure 2 It is a flowchart of another intelligent target recognition method based on multi-source heterogeneous information fusion provided according to an embodiment of the present invention. In this embodiment, based on the above embodiment, optionally, the target recognition model includes a mask region-based convolutional neural network model; based on the fused feature information and the pre-trained target recognition model, the target object information of the target object contained in the target environment is obtained, including: inputting the fused feature information into the pre-trained mask region-based convolutional neural network model to obtain the target object information of the target object contained in the target environment. Among them, the explanations of the terms that are the same or corresponding to the above embodiments are not repeated here. Figure 2 As shown, the method includes:

[0043] S210: Acquire optical images and radar images collected of the target environment.

[0044] S220: extract first feature information corresponding to the optically collected image and second feature information corresponding to the radar collected image, and perform a feature fusion operation on the first feature information and the second feature information to obtain fused feature information.

[0045] Optionally, a method for performing a feature fusion operation on the first feature information and the second feature information to obtain fused feature information includes: performing a feature fusion operation on the first feature information and the second feature information based on a channel attention mechanism to obtain fused feature information.

[0046] Specifically, the process of performing feature fusion operation on the first feature information and the second feature information based on the channel attention mechanism can be as follows: the first feature information and the second feature information are respectively globally averaged pooled to obtain the first global information corresponding to the first feature information and the second global information corresponding to the second feature information, so as to capture the global statistical characteristics of each channel. Then, the first global information and the second global information are processed by the convolution layer to generate the first attention weight corresponding to the first feature information and the second attention weight corresponding to the second feature information, so as to realize the local interaction between the channels; and the first feature information is multiplied by the first attention weight to obtain the first product, and the second feature information is multiplied by the second attention weight to obtain the second product, so as to enhance the feature representation by the first product and the second product. Thereafter, the first product and the second product are input into the ViT (Vision Transformer) encoder for feature-level fusion to obtain the fused feature information.

[0047] It should be noted that the ViT encoder uses the Transformer architecture. First, the input information is divided into fixed-size blocks, and these blocks are converted into vectors for embedding operations, and position encoding is added to maintain spatial information. The encoded sequence is input into the Transformer encoder and processed using a multi-head self-attention mechanism to capture the global features and long-distance dependencies of the image. Finally, the fusion feature information of the first feature information and the second feature information is generated.

[0048] S230, inputting the fused feature information into a pre-trained mask region-based convolutional neural network model to obtain target object information of the target object contained in the target environment.

[0049] Among them, the Mask R-CNN (Region-based Convolutional Neural Network) model is a deep learning model for target detection and image instance segmentation. Exemplarily, the Mask R-CNN model can adopt a deep convolutional neural network based on the residual network (ResNet) architecture, for example, the Mask R-CNN model is a ResNet50 network.

[0050] In this embodiment, after the fused feature information is input into the mask region-based convolutional neural network model, the fused feature information can be processed as follows: the fused feature information is input into the multi-scale feature pyramid network to output a fused feature map corresponding to the fused feature information; the fused feature map is input into the region proposal network to obtain a detection frame corresponding to the fused feature map; the detection frame and the fused feature map are calibrated with the region of interest (ROI Align) operation, and the target object information of the target object is determined based on the result obtained after the ROI alignment operation. Among them, the detection frame is the predicted area containing the target object in the fused feature map.

[0051] It should be noted that this embodiment uses the detection head designed in the region proposal network to perform visual target recognition, and generates at least one detection frame on the fused feature map by designing a preset frame diagram. The core operation of the region proposal network is to generate a series of anchor frames according to a certain rule based on the fused feature map. After obtaining the anchor frame, it is determined whether there is an object to be identified in the anchor frame, and the anchor frame is adjusted to make it closer to the true value. The region proposal network generates detection frames quickly and can achieve predictions quickly and accurately.

[0052] This embodiment uses a multi-scale feature pyramid network to perform a deconvolution operation on a layer with a small size but more feature information, and fuses the result of the deconvolution operation with a layer with a large size but less feature extraction to obtain a fused feature map. The multi-scale feature pyramid network can detect target objects at different scales, which is beneficial to improving the accuracy and robustness of the target detection results. In addition, bilinear interpolation is performed inside each region of interest on the fused feature map through the ROI Align operation, and the generated detection frame is mapped to the fused feature image, which can effectively solve the quantization error problem.

[0053] The above describes in detail the embodiments corresponding to the intelligent target recognition method based on multi-source heterogeneous information fusion. In order to make the technical scheme of the present method further clear to those skilled in the art, a specific application scenario is given below.

[0054] In this embodiment, optical sensors and synthetic aperture radars may be deployed on an aircraft, the synthetic aperture radar may be used to collect radar images of the target environment, and the optical sensors may be used to collect optical images. Figure 3 is a schematic diagram of the working process of an intelligent target recognition device based on multi-source heterogeneous information fusion provided according to an embodiment of the present invention; Figure 3As shown, the optical image acquisition module can be deployed on the optical sensor to collect images of the target environment based on optical imaging technology. The SAR image acquisition module is deployed on the synthetic aperture radar, and uses the target echo signal received by the synthetic aperture radar to generate the radar acquisition image of the target environment. The optically acquired image is processed by the optical image preprocessing module, and the image geometry transformation, color space adjustment, noise addition, and image cropping are performed on the optically acquired image. The radar acquired image is filtered by the SAR image preprocessing module, and the detailed features of the radar acquired image are increased by combining contrast enhancement, histogram equalization, and transposition.

[0055] The preprocessed optical acquisition image is input into the optical image feature extraction module to extract the first feature information corresponding to the optical acquisition image. The preprocessed radar acquisition image is input into the SAR image feature extraction module to obtain the second feature information corresponding to the radar acquisition image. The first feature information and the second feature information are subjected to feature fusion processing through the optical-SAR image feature fusion module to obtain fused feature information. The fused feature information is passed through the target recognition module to identify the region information and category information of the target object contained in the target environment.

[0056] This embodiment performs multi-source heterogeneous information fusion for optically collected images and radar images collected by synthetic aperture radar, designs a feature-level fusion model based on deep neural network, effectively fuses features, gives full play to the strong penetration characteristics of synthetic aperture radar all day and all weather, combines the rich spectral features contained in optically collected images, forms information complementary advantages, is conducive to accurate target recognition operations, and effectively improves the accuracy and generalization of target recognition in complex scenarios.

[0057] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0058] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent target recognition method based on multi-source heterogeneous information fusion, It is characterized in that include: Acquire an optical acquisition image and a radar acquisition image acquired of the target environment; wherein the radar acquisition image is an image generated by a synthetic aperture radar based on a received target echo signal corresponding to the target environment; Extracting first feature information corresponding to the optically collected image and second feature information corresponding to the radar collected image, and performing a feature fusion operation on the first feature information and the second feature information to obtain fused feature information; Based on the fused feature information and a pre-trained target recognition model, target object information of the target object contained in the target environment is obtained.

2. The method according to claim 1, It is characterized in that Before extracting the first feature information corresponding to the optically collected image, the method further includes: Performing at least one image preprocessing operation of image geometric transformation, color space adjustment, noise addition and image cropping on the optically collected image; The optically acquired image is updated based on the image obtained after the image preprocessing operation.

3. The method according to claim 2, It is characterized in that The image geometric transformation includes at least one of image rotation, size adjustment, horizontal flipping and vertical flipping; the color space adjustment includes at least one of brightness adjustment, contrast adjustment and saturation adjustment; the noise addition includes adding Gaussian noise and / or adding salt and pepper noise; the image cropping includes random cropping and / or center cropping.

4. The method according to claim 1, It is characterized in that Before extracting the second feature information corresponding to the radar collected image, the method further includes: Perform detail feature enhancement processing on the radar collected image, and update the image obtained after the detail feature enhancement processing as the radar collected image.

5. The method according to claim 4, It is characterized in that The detail feature enhancement process includes at least one of a contrast enhancement process, a histogram equalization process and a transposition process.

6. The method according to claim 1, It is characterized in that Before extracting the second feature information corresponding to the radar collected image, the method further includes: Based on a spatial adaptive filtering algorithm, a filtering preprocessing operation is performed on the radar collected image, and the image obtained after the filtering preprocessing operation is updated as the radar collected image.

7. The method according to claim 1, It is characterized in that The extracting the first feature information corresponding to the optically collected image and the second feature information corresponding to the radar collected image includes: Extracting first feature information of the optically collected image based on a pre-trained first feature extraction model; Based on a pre-trained second feature extraction model, second feature information of the radar acquired image is extracted.

8. The method according to claim 7, It is characterized in that The first feature extraction model and the second feature extraction model include a convolutional neural network model and / or a deep learning model of a self-attention mechanism.

9. The method according to claim 1, It is characterized in that The target recognition model includes a mask region-based convolutional neural network model; The step of obtaining target object information of the target object contained in the target environment based on the fused feature information and the pre-trained target recognition model includes: The fused feature information is input into the pre-trained mask region-based convolutional neural network model to obtain target object information of the target object contained in the target environment.

10. The method according to claim 1, It is characterized in that The performing a feature fusion operation on the first feature information and the second feature information to obtain fused feature information includes: Based on the channel attention mechanism, a feature fusion operation is performed on the first feature information and the second feature information to obtain the fused feature information.

Citation Information

Patent Citations

  • Airport runway foreign matter fusion detection method based on multi-source heterogeneous sensor

    CN114137511A

  • Millimeter wave radar and vision fused three-dimensional target detection method based on attention mechanism

    CN114708585A

  • Multi-modal fusion target detection method of mixed attention mechanism

    CN116704304A

  • Radar target semantic information rapid labeling method, system and device

    CN116908852A

  • Coal rock identification method based on multi-modal cone point cloud fusion

    CN117636034A

Cited By

  • Ship image classification device based on deep neural network feature coding

    CN120070953A