Remote sensing target detection method based on centralized visual processing center

Abstract

The invention relates to the technical field of remote sensing and computer vision, and discloses a remote sensing target detection method based on a centralized visual processing center, the centralized visual processing center (CVPC) is firstly provided, the structure is a visual processing center in which a Tranformer encoder and a CNN are parallel, the Light encoder is used for capturing a global long-range dependency relationship, a Pixel level learning center (PLC) module is used for establishing a pixel-level correlation, and the CNN is used for establishing a pixel-level correlation relationship; and the representation capability of detail features is enhanced. The CVPC effectively improves the detection efficiency of remote sensing targets with high feature similarity and serious shielding. Secondly, a centralized feature cross-layer fusion pyramid structure is provided, fusion with a result of a centralized visual processing center is carried out from top to bottom, and the detail expression ability of each layer of features is enhanced. Finally, a sparse convolution detection head (CEASC) is introduced, and the detection efficiency is ensured while the accuracy is improved. According to the 3CNet provided by the invention, on the premise of balancing the detection speed, the relatively advanced detection precision (DOTA-v1.0 is 78.62% mAP, the DIOR is 79.12% mAP, and the RSOD is 95.50% mAP) is realized.

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing and computer vision, and in particular to a remote sensing target detection method based on a centralized vision processing center. Background Art

[0002] The rapid development of satellite sensors and remote sensing technology has greatly promoted the research of remote sensing image target detection. The high efficiency of remote sensing image target detection has made it widely used in many fields, such as urban monitoring, forest early warning, resource detection, crop monitoring, etc. Remote sensing images have the characteristics of large target scale span, dense distribution, high feature similarity, and severe occlusion, which makes it difficult for existing detection models to achieve the ideal effect of balancing detection accuracy and efficiency.

[0003] Against the backdrop of the rapid development of deep learning, object detection tasks based on convolutional neural networks (CNNs) have made significant progress. However, most object detection models are designed for object detection tasks in conventional scenarios, and are trained and tested using the MSCOCO dataset as a benchmark. If these object detection models used in conventional scenarios are directly applied to remote sensing image object detection, there will be two major problems: first, the size span of remote sensing targets is too large and the distribution is dense, which leads to the occurrence of missed detection problems in conventional detection models; second, remote sensing overhead imaging, wide field of view, and complex background lead to high similarity of target features and severe occlusion, resulting in a high false detection rate in conventional detection models. CNN-based object detection models perform downsampling during feature extraction to reduce the amount of calculation, which can easily lead to feature blurring or even loss of small target features. When target features and background features have similar textures and sizes, they are difficult to distinguish. In addition, occlusion problems usually lead to incomplete feature learning, resulting in semantic ambiguity. Therefore, more refined spatial features and global context information are needed as clues for target feature learning.

[0004] CNN has the advantage of spatial position expression, but due to the locality of convolution operations, it is difficult to directly model global semantic interactions and contextual information. Researchers have made different attempts to address this issue. For example, FPN uses the idea of ​​image pyramids and proposes a top-down inter-layer feature interaction structure that enables shallow features to obtain the semantic representation of deep features. ExFuse proposes to introduce semantic information into low-level features, while introducing shallow high-resolution details into high-level features to enhance the later feature fusion effect. Lin uses the inherent multi-scale pyramid hierarchy of deep convolutional networks and adds branches to construct feature pyramids at marginal additional costs. These methods do not directly encode global contextual information, and it is difficult to directly obtain clear global scene information from remote sensing images with complex backgrounds.

[0005] To address the limited receptive field problem of CNN, recently proposed methods for object detection based on visual transformers. These methods first divide the input image into different image blocks, and then use the feature interaction between blocks based on multi-head attention to obtain global long-range dependencies. Although these methods can enhance the utilization of limited receptive field and local context information in CNN, they also increase the computational complexity. Research on other advanced methods shows that shallow features mainly contain common feature patterns of objects, such as texture, color, and orientation, which are often not global; in contrast, deep features reflect object-specific information, which usually requires global information. Therefore, we believe that transformer encoders can be used in deep features. The Convolutional Multi-Head Self-Attention (CMHSA) of the Efficient Convolutional Transformation Block (ECTB) proposed by Tao et al. uses convolutional projection instead of positional linear projection to extract the contextual information of the object to improve the recognition ability of occluded objects, while reducing the amount of computation without significantly improving the accuracy of occluded object detection.

[0006] Remote sensing image target detection is widely used in various fields, and many people have studied its characteristics. Ouyang et al. modified the YOLOv3 network structure so that the small feature map corresponding to the smaller target can also be fused with the feature map of the large target, realizing the simultaneous detection of large and small targets. Lu et al. designed an end-to-end network of attention feature fusion SSD, using spatial attention and channel attention to suppress background noise and highlight key features, solving the problem of complex background in remote sensing target detection. Chen et al. used the model of natural images to propose the image-level and strength-level structure of Faster R-CNN with domain adaptation to solve the problem of insufficient remote sensing image data. Yang et al. used the CNN model to realize the dual detection of position and direction of remote sensing images. The above methods make full use of the advantages of CNN in spatial position to solve the problems of multi-scale and complex background of remote sensing image targets. However, the high feature similarity and serious occlusion problems caused by complex backgrounds have not been well solved.

[0007] Attention mechanisms are widely used in semantic segmentation and image detection tasks. When using attention to process tasks, the importance of different information is reflected by weights. The channel attention mechanism SENet is divided into two parts: compression and excitation, which enables it to adaptively select and emphasize important features and improve the expressiveness of features. The spatial attention mechanism STN can transform various deformation data in space and automatically capture important regional features. The hybrid attention mechanism CBAM applies channel-level attention and spatial-level attention to adaptive feature refinement. The self-attention mechanism ASPP uses dilated convolutions with different dilation rates to perform convolution operations on the input feature map to obtain the contextual information of the image at multiple different scales.

[0008] Transformer was first proposed for machine translation tasks and surpassed previous sequence transformation models based on complex recursion or CNN. Subsequently, Transformer-based models have been widely used in various fields of natural language processing. The standard Transformer block consists of multi-head self-attention (MSA), multi-layer perceptron (MLP), and layer normalization (LN). In order to alleviate the large amount of computation of complex Transformer models, PVT introduces a pyramid structure into ViT to obtain multi-scale feature maps. Specifically, it flexibly controls the length of the Transformer sequence through the patch embedding layer. Although PVT reduces the consumption of computational resources to a certain extent, its complexity is still quadratic to the image size. Liu et al. proposed the Swin Transformer based on the shifted window strategy, which limits the computation of MSA to non-overlapping windows while allowing cross-window information interaction. The Swin Transformer has only linear computational complexity and achieves advanced performance in various visual tasks, including image classification, object detection, and semantic segmentation. Although Transformer models perform well in computer vision tasks and can better establish long-range dependencies / global relationships of features, they still lack the ability to capture detailed feature expressions, resulting in suboptimal detection accuracy when dealing with small targets and complex backgrounds. Summary of the invention

[0009] In order to solve the technical problems of high similarity and serious occlusion of target features in remote sensing images, the present invention provides a remote sensing target detection method based on a centralized visual processing center. While adopting the advantages of CNN in spatial position, a new framework is proposed using the Transformer method to inherit the advantages of both to solve the problems of high feature similarity and serious occlusion; multiple average pooling layers are used to extract vertical and horizontal channel dependencies, and then pixel-level correlations are established by multiplying their features to enhance the representation capability of detail features; Lightweight encoder is used to capture the global context information and long-range dependencies of the input image, and a PLC module is combined to capture fine-grained feature representation, so as to improve the accuracy of remote sensing target detection under complex backgrounds.

[0010] The remote sensing target detection method based on a centralized visual processing center provided by the present invention comprises the following steps:

[0011] a. Build a centralized visual processing center (CVPC) that includes a Transformer encoder and a convolutional neural network (CNN) in parallel, where;

[0012] b. Construct a centralized feature cross-layer fusion pyramid structure and fuse it with the results of the centralized visual processing center (CVPC) to enhance the detail expression ability of each layer of features;

[0013] c. Introduce sparse convolution detection head (CEASC) to improve detection accuracy while ensuring detection efficiency.

[0014] Preferably, the centralized visual processing center (CVPC) includes a Lightweight encoder and a Pixel level Learning Center (PLC) module. The Lightweight encoder is used to capture global long-range dependencies, and the Pixel level Learning Center (PLC) module is used to establish pixel-level correlations to enhance the representation capability of detail features.

[0015] Preferably, the Lightweight encoder module consists of two residual blocks, namely a neighborhood attention module and a channel MLP-based module. The Transformer encoder of the centralized visual processing center (CVPC) adopts the Lightweight encoder, which can reduce the amount of calculation and improve the detection efficiency.

[0016] Preferably, the pixel level learning center (PLC) module establishes pixel-level correlation through global average pooling and matrix multiplication in the spatial direction to enhance the detection capability of occluded targets in remote sensing images.

[0017] Preferably, the centralized feature cross-layer fusion pyramid structure fuses high-level semantic information with low-level detail information in a top-down manner to improve the accuracy of remote sensing target detection.

[0018] Preferably, the sparse convolution detection head (CEASC) adopts an adaptive multi-layer mask strategy to generate an optimal mask ratio. The sparse convolution detection head (CEASC) improves the accuracy of target detection while maintaining detection efficiency through sparse convolution operations.

[0019] Compared with the related art, the remote sensing target detection method based on the centralized visual processing center provided by the present invention has the following beneficial effects:

[0020] The present invention provides a remote sensing target detection method based on a centralized visual processing center:

[0021] 1. A centralized visual processing center (CVPC) is proposed. This structure is a visual processing center that parallels the Transformer encoder and CNN. The Lightweight encoder is used to capture the global long-range dependencies, and the Pixellevel learning center (PLC) module is used to establish pixel-level correlations, enhance the representation of detail features, and improve the detection accuracy of remote sensing targets under complex backgrounds.

[0022] 2. A centralized feature cross-layer fusion pyramid structure is proposed, which is integrated with the results of the centralized visual processing center in a top-down manner, enhancing the detail expression ability of each layer of features and improving the detection accuracy of remote sensing targets at all scales;

[0023] 3. The sparse convolution detection head (CEASC) was introduced to improve accuracy while ensuring detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is the overall structure diagram of CF-YOLO;

[0025] Figure 2 This is the CVPC structure diagram;

[0026] Figure 3 This is a pyramid structure diagram for centralized feature cross-layer fusion;

[0027] Figure 4 This is the CEASC structure diagram;

[0028] Figure 5 Visualization of CF-YOLO detection results in the DOTA-v1.0 dataset;

[0029] Figure 6 Visualization of CF-YOLO detection results in the DIOR dataset;

[0030] Figure 7 Visualization of CF-YOLO detection results in the RSOD dataset. DETAILED DESCRIPTION

[0031] The present invention will be further described below in conjunction with the accompanying drawings and implementation modes.

[0032] like Figure 1 As shown in FIG. 1 , this is the overall structure diagram of CF-YOLO in the present invention. The upper part is the overall structure of the network, and the lower part is the details of each operation.

[0033] The invention proposes a remote sensing target detection method based on a centralized visual processing center, comprising the steps of: constructing a centralized visual processing center (CVPC); constructing a centralized feature cross-layer fusion pyramid structure; and introducing a sparse convolution detection head (CEASC).

[0034] A. Centralized Vision Processing Center (CVPC):

[0035] We take advantage of CNN’s spatial position and visual Transformer’s global context information and propose a centralized visual processing center (CVPC). CVPC aims to effectively improve the detection efficiency of remote sensing targets with high feature similarity and severe occlusion. Figure 2 As shown in the figure, we proposed that CVPC is mainly composed of two parallel connected blocks, in which the Lightweight encoder is used to capture the global long-range dependencies; the PLC module is used to establish pixel-level correlation and the representation ability of strong detail features. The resulting feature maps of these two blocks are connected together along the channel dimension as the input of the downstream centralized feature cross-layer fusion pyramid. The above process can be expressed as:

[0036] X=cat(LC(X in ), PLC(X in ))

[0037] Where X is the output of CVPC; cat(.) represents the operation of connecting feature maps along the channel dimension; LC(X in )、PLC(X in ) represent the output features of Lightweight encoder and PLC module respectively. in is the output feature of the Backbone bottom after the SPPCSPC operation. Lightweight encoder: The Lightweight encoder is mainly composed of two residual blocks: the neighborhood attention module and the channel MLP-based module. The input of the channel MLP-based module is the output of the neighborhood attention module. Both blocks are followed by Dropout operations to help the network converge better, prevent network overfitting, and improve feature generalization and robustness. For the multi-head neighborhood attention module, first convert X in Feed it to the LayerNorm layer, then implement the NeighborhoodAttention operation on these features, then perform the Dropout operation, and finally add it to X in Add to realize the residual structure. The above process is expressed as:

[0038]

[0039] in is the output of the neighborhood attention module, LN(.) is the LayerNorm operation, and NA(·) is the NeighborhoodAttention operation.

[0040] For the channel MLP-based module, we first convert the output features of the multi-head neighborhood attention module into

[0041] The features are fed to the LayerNorm layer, and then channel MLP is implemented on these features. Compared with spatial MLP, channel MLP can effectively reduce the computational complexity. Then the Dropout operation and Connection. The above process is expressed as:

[0042]

[0043] Where LN(.) is the LayerNorm operation and CMLP(.) is the channel MLP. PLC: Given a feature map X∈R (h×w)×c1 ,In order to reduce the computational cost, the number of feature map channels is reduced to c1 / 2 through the Reshape operation. Then the feature X is fed into a 3×3 dilated convolution layer with a dilation rate of 2 to reconstruct the structural information of the feature map through a large receptive field. The global average pooling operation is used for the vertical and horizontal spatial directions, and the calculation formula of the elements in each direction is expressed as:

[0044]

[0045] Where i, j, q are the indices of the vertical direction, horizontal direction and channel respectively, 0≤i<h, 0≤j<w, 0≤q<c1 / 2. BN(.) is BatchNormalization, and GELU(.) is the dilated convolution layer with GELU activation function. We get the aggregation tensor k in the vertical and horizontal directions from the above formulas h and k w , where k h ∈R (h×1)×c1 / 2 , k w ∈R (1×w)×c1 / 2 . k h and k w The pixel-level weights of the feature map converge spatially, so we perform matrix multiplication on the two to obtain the position-related feature map M, M∈R (h×w)×c1 / 2 Finally, M is passed through a 1×1 convolution layer and reshaped again, and the number of channels is restored to C1 to obtain the output feature map PLC (X in ),PLC(X in )∈R (h×w)×c1 , the calculation process is expressed as follows:

[0046]

[0047] Where × represents matrix multiplication, Represents a 1×1 convolutional layer with Batch Normalization and GELU.

[0048] B. Centralized feature cross-layer fusion pyramid:

[0049] The Backbone network is used for feature extraction. As the number of network layers increases, the network's receptive field becomes larger and larger, and its semantic expression ability is enhanced. However, after multiple layers of convolution operations, many detailed features become blurred or even lost. Using the deep feature maps of the Backbone for downstream target detection tasks may lead to a decrease in the model's detection ability. The centralized visual processing center (CVPC) is used to centralize feature expression and enhance the representation of detailed features. Therefore, we propose a centralized feature cross-layer fusion pyramid structure to fuse the result feature map of the centralized visual processing center (CVPC) with the Neck part. Figure 3 As shown in the figure, we added four branches, which upsampled the result feature map of the centralized visual processing center (CVPC) and then concat the features with Stage1, Stage2, Stage3, and Stage4. The centralized feature cross-layer fusion pyramid effectively integrates the context information of the CVPC and Neck detection layers, provides more comprehensive and rich visual context information for the Head, improves the detection accuracy of remote sensing targets at all scales, and makes the model more robust and accurate.

[0050] C. Sparse Convolutional Detection Head (CEASC):

[0051] While the cross-layer fusion of multi-scale features brings the benefit of improving accuracy, it may also bring higher computational complexity. Optimizing computational efficiency and reducing computational complexity is a problem that needs to be solved. The present invention introduces an optimized design of a detection head based on sparse convolution (CEASC), which effectively balances detection accuracy and efficiency. CEASC proposes a new global context-enhanced adaptive sparse convolutional network. The algorithm first develops a context-enhanced group normalization (CE-GN) layer to replace the statistics based on sparse sampling features with global context features, and then designs an adaptive multi-layer mask strategy to generate optimal mask ratios at different scales to achieve compact foreground coverage and improve accuracy and efficiency. The CEASC detection head structure is as follows: Figure 4 shown.

[0052] To verify the effectiveness and robustness of our proposed CF-YOLO method, we conducted extensive experiments on three large-scale public remote sensing datasets: DOTA-v1.0, DIOR, and RSDO, and compared them with advanced methods.

[0053] A.Dataset:

[0054] The DOTA dataset is divided into three versions: v1.0, v1.5 and v2.0. Here, we use the DOTA-v1.0 version. DOTA-v1.0 contains 2806 images and 188282 instances. The image size ranges from 800×800 to 4000×4000. The objects are divided into 15 categories, namely: helicopter (HC), port (HA), ship (SH), large vehicle (LV), tennis court (TC), airplane (PL), small vehicle (SV), bridge (BR), baseball field (BD), ground track and field (GTF), basketball court (BC), roundabout (RA), football field (SBF), storage tank (ST) and swimming pool (SP). The training, validation and test set split ratio of this dataset is 3:1:2.

[0055] The DIOR dataset contains 23,463 images and 192,518 instances. The image size is fixed at 800×800. The objects are divided into 20 categories, namely chimney (CH), airplane (APL), train station (TS), airport (APO), basketball court (BC), bridge (BR), storage tank (STO), dam (DAM), highway service area (ESA), baseball field (BF), highway toll station (ETS), golf course (GF), seaport (HA), overpass (OP), ship (SH), windmill (WM), ground track and field (GTF), tennis court (TC), vehicle (VE) and stadium (STA). The training and validation sets contain 11,725 ​​images and 68,073 instances, and the test set contains 11,738 images and 124,445 instances.

[0056] The RSOD dataset contains four types: airplanes, fuel tanks, playgrounds, and overpasses, and is annotated in the format of the PASCAL VOC dataset. The dataset includes 4 folders, each containing an object: airplane dataset, 4993 airplane instances in 446 images; playground, 191 playground instances in 189 images; overpass, 180 overpass instances in 176 images; fuel tank, 1586 fuel tank instances in 165 images. The training, validation, and test set split ratios of this dataset are 3:1:2.

[0057] B. Implementation details

[0058] (1) Baseline: In order to verify the universality of CVPC and the applicability of the overall idea in remote sensing target detection, we selected YOLOv7 as the baseline, used an end-to-end training strategy, and followed the default training and inference settings of YOLOv7. Unlike the previous YOLO series models, YOLOv7 introduced model reparameterization into the network architecture. The idea of ​​reparameterization first appeared in REPVGG; a training method for auxiliary heads was proposed to increase training costs and improve accuracy, but it does not affect the inference time because the auxiliary heads only appear during training. In addition, the YOLOv7 label assignment strategy adopts the cross-grid search of YOLOV5 and the matching strategy of YOLOX.

[0059] (2) Training Settings: We implement our network on the public deep learning framework PyTorch. All experiments are conducted on an A6000 GPU with the following experimental parameters: epoch = 300, b atchsize =16, imgsize=640×640.

[0060] (3) Evaluation metrics: In the ablation experiment, we mainly followed the commonly used target detection evaluation metric - mean average precision (mAP@0.5). In addition, we also used three parameters, Prams (M), GFLOPs (G), and Speed ​​(ms), to quantify the efficiency of the model. In the comparative experiment, only the mean average precision (mAP@0.5) metric was used.

[0061] C.Ablation study

[0062] We used the DOTA-V1.0 dataset to conduct ablation experiments to compare the effects of the centralized visual processing center (CVPC), centralized feature cross-layer fusion pyramid, and sparse convolutional detection head (CEASC).

[0063] (1) Effect of centralized visual processing center (CVPC): In remote sensing images, the features of the "large and small vehicles" category are easily similar to complex backgrounds and easily blocked by shadows and trees, making it difficult to extract and recognize its semantic features. Therefore, we specifically compared the detection effects of "large and small vehicles". As shown in Table I, in order to verify the effectiveness of Lightweight encoder and PLC, we conducted an ablation experiment on the baseline. After using the Lightweight encoder module alone, the detection accuracy of "large and small vehicles" increased by 2.01% and 2.23% respectively, and the average detection accuracy of all types increased by 1.14%. After using the PLC module alone, the detection accuracy of "large and small vehicles" increased by 2.08% and 2.30% respectively, and the average detection accuracy of all types increased by 1.08%. After using the Lightweight encoder module and the PLC module at the same time, the detection accuracy of "large and small vehicles" increased by 4.12% and 3.98% respectively, and the average detection accuracy of all types increased by 1.88%. It can be seen that the introduction of CVPC effectively reduces the negative impact of high target feature similarity and severe occlusion caused by the complex background of remote sensing images.

[0064] Table I Comparison of CVPC ablation experiments

[0065]

[0066] (2) Effect of centralized feature cross-layer fusion pyramid: In order to verify the effectiveness of centralized feature cross-layer fusion pyramid, we used four-stage feature fusion schemes for ablation experiments. In the first stage, only the results of CVPC were fused with Stage 1, and then one stage of feature fusion was added in sequence to test the effect. As shown in Table II, it can be seen that as the stage of fusion of CVPC results with the following hierarchical features increases, the detection effect becomes better. It can be seen that the introduction of centralized feature cross-layer fusion pyramid effectively increases the accuracy of remote sensing target detection, while having little effect on computational efficiency.

[0067] Table II Comparison of centralized feature cross-layer fusion pyramid ablation experiments

[0068]

[0069]

[0070] (3) Effect of sparse convolution detection head (CEASC): In order to verify the effectiveness of the CEASC detection head, we conducted ablation experiments on the baseline, CEASC and DyHead. As shown in Table III, we used CEASC and DyHead to replace the original YOLOv7 detection head respectively. It can be seen that CEASC effectively balances the detection accuracy and efficiency. While ensuring the detection accuracy, the sparse convolution is used to greatly reduce the operation parameters and improve the detection efficiency.

[0071] Table III Comparison of CEASC ablation experiments

[0072]

[0073] B. Comparisons With State-of-the-Arts

[0074] (1) Experimental results of DOTA dataset: We conducted experiments on the HBB task of the DOTA-v1.0 dataset. We patch-cropped the images in this dataset to expand the data volume and solve the problem of small category distribution. This remote sensing dataset has the problems of large target scale span, dense distribution, high feature similarity, and severe occlusion, but our proposed method can solve these challenging problems well and has achieved good results in all scale categories. At the same time, it is particularly effective for the "large and small vehicles" category that is similar to complex background features and easily obscured by shadows and trees. As can be seen from Table IV, LV and SV have the largest accuracy improvements, which are 3.46% and 3.04% respectively. Figure 5 As shown in Figure 2, we visually compare the model effects.

[0075] Table IV Comparative experiments on the DOTA-v1.0 dataset

[0076]

[0077] (2) Experimental results on the DIOR dataset: We used the DIOR dataset to conduct comparative experiments with several advanced horizontal box detectors under the same conditions as the validation of the DOTA dataset. The data results are shown in Table V. All experiments use the train and val sets in training and are evaluated on the test set. The experimental results show that our method also achieves good results in various scale categories on this dataset. At the same time, it is particularly effective for the "vehicle" category that is similar to complex background features and easily blocked by shadows and trees. From Table V, we can see that VE improves the accuracy the most, by 7.01%. This verifies the robustness and generalization of our method, and our CF-YOLO network is challenging for recent horizontal detectors and has a large room for growth. Figure 6 A comparison chart of the effects is shown.

[0078] Table V Comparative experiments on the DIOR dataset

[0079]

[0080]

[0081] (3) Experimental results on RSOD dataset: We conducted comparative experiments with several advanced horizontal box detectors using the RSOD dataset under the same conditions as the validation of the DOTA dataset, and listed their results in Table VI. All experiments use the train and val sets in training and are evaluated on the test set. Figure 7 A comparison chart of the effects is shown.

[0082] Table VI Comparative experiments on RSOD dataset

[0083]

[0084] Compared with the related art, the remote sensing target detection method based on the centralized visual processing center provided by the present invention has the following beneficial effects:

[0085] In the present invention, we first propose a centralized visual processing center (CVPC), which is a visual processing center with a Transformer encoder module and a CNN module in parallel, in which the Lightweight encoder is used to capture the global long-range dependency, and the PLC module is used to establish pixel-level correlation, enhance the representation ability of detail features, and improve the detection accuracy of remote sensing targets under complex backgrounds; CVPC effectively improves the detection efficiency of remote sensing targets with high feature similarity and severe occlusion. Secondly, the present invention proposes a centralized feature cross-layer fusion pyramid structure, which is fused with the results of the centralized visual processing center in a top-down manner, enhances the detail expression ability of each layer of features, and improves the detection accuracy of remote sensing targets at all scales. Finally, we introduce a sparse convolution detection head (CEASC) to improve accuracy while ensuring detection efficiency. The present invention uses YOLOv7 as the baseline and has achieved considerable detection performance on three public remote sensing datasets: DOTA-v1.0, DIOR, and RSOD.

[0086] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A remote sensing target detection method based on a centralized visual processing center, It is characterized in that The following steps are involved: a. Build a centralized visual processing center (CVPC) that includes a Transformer encoder and a convolutional neural network (CNN) in parallel, where; b. Construct a centralized feature cross-layer fusion pyramid structure and fuse it with the results of the centralized visual processing center (CVPC) to enhance the detail expression ability of each layer of features; c. Introduce sparse convolution detection head (CEASC) to improve detection accuracy while ensuring detection efficiency.

2. The remote sensing target detection method based on a centralized visual processing center according to claim 1, It is characterized in that The centralized visual processing center (CVPC) includes the Lightweight encoder and the Pixel level Learning Center (PLC) modules. The Lightweight encoder captures the global long-range dependencies, and the Pixel level Learning Center (PLC) module establishes pixel-level correlations to enhance the representation capability of detail features.

3. The remote sensing target detection method based on a centralized visual processing center according to claim 2, It is characterized in that The Lightweight encoder module consists of two residual blocks, namely the neighborhood attention module and the channel MLP-based module.

4. The remote sensing target detection method based on a centralized visual processing center according to claim 2, It is characterized in that The pixel level learning center (PLC) module establishes pixel-level correlations through global average pooling and matrix multiplication in the spatial direction to enhance the detection capability of occluded targets in remote sensing images.

5. The remote sensing target detection method based on a centralized visual processing center according to claim 1, It is characterized in that The centralized feature cross-layer fusion pyramid structure adopts a top-down approach to fuse high-level semantic information with low-level detail information to improve the accuracy of remote sensing target detection.

6. The remote sensing target detection method based on a centralized visual processing center according to claim 1, It is characterized in that The sparse convolution detection head (CEASC) adopts an adaptive multi-layer mask strategy to generate the optimal mask ratio. The sparse convolution detection head (CEASC) performs sparse convolution operations to improve the accuracy of target detection while maintaining detection efficiency.

Images