Improved YOLOv4 network model and small target detection method

Abstract

According to the improved YOLOv4 network model and the small target detection method, the detection precision of the small target can be effectively improved on the premise of ensuring the detection speed. The backbone structure network module is used for carrying out feature extraction on a target image and outputting multiple layers of feature maps with different scales from shallow to deep, and the feature fusion module is used for carrying out feature enhancement on the feature maps layer by layer from deep to shallow and respectively and correspondingly splicing the feature maps with feature maps with the same scale after down-sampling from shallow to deep to obtain a fused feature map; the feature fusion module comprises a feature enhancement module and a high-efficiency double attention module, the feature enhancement module is used for expanding the receptive field of the feature map and enhancing semantic information of an up-sampling feature map in a bottom-up path, and the high-efficiency double attention module is used for enhancing target response and inhibiting background interference. The feature map participating in feature fusion comprises a shallow-layer feature map containing bottom-layer features after quadruple downsampling, and the multi-classifier module is used for carrying out classification detection on the fused feature map and then outputting a detection result.

Description

technical field [0001] The invention relates to the technical field of target detection, in particular to an improved YOLOv4 network model and a small target detection method. Background technique [0002] At present, target detection algorithms based on deep learning can generally be divided into two-stage detection algorithms and one-stage detection algorithms. Two-stage detection algorithms, such as Faster R-CNN, first use the region proposal network to generate candidate regions, and then classify and regress the candidate regions to obtain the final detection result. A one-stage detection algorithm, such as YOLO, divides the input image into S*S grids, each grid is responsible for the detection of the target centered on the grid, and predicts the bounding box, location reliability and target belonging to each grid. The probability of the category is finally obtained through non-maximum suppression to obtain the final detection result. The two types of target detection...

AI Technical Summary

Problems solved by technology

However, it ignores the semantic conflicts caused by the direct fusion of information of different densities, resulting in tiny information that may be overwhelmed by conflicting information

In addition, the context information is not fully utilized, and the lack of feature enhancement capabilities limits the further improvement of the model's detection accuracy for small targets.

Images