Multi-scale and attention mechanism combined remote sensing image aircraft target detection method
A remote sensing image and aircraft target technology, applied in the field of remote sensing image aircraft target detection, can solve problems such as loss of aircraft feature information, feature information cannot accurately describe the target, etc., and achieve the effects of good detection effect, improved average accuracy, and high detection accuracy.
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0024] Such as figure 2 As shown, the embodiment of the present invention provides an improved YOLO V4 network (referred to as the M-YOLO V4 network in the present invention) for remote sensing image aircraft target detection, and the improved YOLO V4 network uses the 104× output of the original backbone network. 104 feature maps, as the first detection scale; the 104×104 feature maps are 2 times downsampled and fused with the 52×52 feature maps output by the original FPN network (multi-scale feature fusion network) to obtain new 52×52 features Figure, as the second detection scale; the new 52×52 feature map is 2 times downsampled and fused with the 26×26 feature map output by the original FPN network to obtain a new 26×26 feature map, which is used as the third detection scale; The new 26×26 feature map is 2 times downsampled and fused with the 13×13 feature map output by the original FPN network to obtain a new 13×13 feature map as the fourth detection scale; wherein, the i...
Embodiment 2
[0029]In general, the feature map extracted by CNN contains rich target feature information, and there is also a large amount of background information, which will reduce the detection performance. The original YOLO V4 network uses the concat operation to fuse feature maps of different scales, but the concat operation is simply connected in the channel dimension, which cannot reflect the importance and relevance of different features, and the fused feature map cannot accurately describe Target.
[0030] Therefore, on the basis of the above examples, as image 3 As shown, the improved YOLO V4 network provided by the embodiment of the present invention also includes an attention mechanism module, which uses the Squeeze-and-ExcitationBlock (SE Block for short) in SENet; the SE Block is used for feature fusion , assigning weights to each pixel in the channel dimension.
[0031] Specifically, SE Block includes three parts: Squeeze, Excitation and Reweight, such as Figure 4 show...
Embodiment 3
[0035] Based on the improved YOLO V4 network in the above embodiments, the embodiment of the present invention provides a remote sensing image aircraft target detection method that combines multi-scale and attention mechanisms, including:
[0036] S101: Acquire feature maps of four detection scales of the input image, where the four detection scales are respectively a first detection scale, a second detection scale, a third detection scale, and a fourth detection scale;
[0037] S102: Perform seven convolutions on the feature maps of the four detection scales, and then input the convolved feature maps to corresponding SE Blocks to obtain fused feature maps.
[0038] The remote sensing image aircraft target detection method combining multi-scale and attention mechanism provided by the present invention is based on the YOLOV4 algorithm, and the M-YOLO V4 algorithm suitable for combining multi-scale and attention mechanisms is designed for remote sensing image aircraft target dete...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More