Underwater group target detection method fusing YOLOv4 and deformable convolution

Abstract

The invention discloses an underwater group target detection method fusing YOLOv4 and deformable convolution, which is characterized in that a convolutional neural network YOLOv4 and the deformable convolution are fused, adaptive threshold output is added to serve as a network model, the offset of sampling points is learned through the deformable convolution, the sampling points are moved to a target, and the target is detected. According to the method, the problem of underwater image blurring can be relieved without image restoration, image enhancement and a complex network structure, and meanwhile, the problem of inaccurate positioning caused by spatial offset is relieved by splicing original input of the module and an output result with offset information. According to the method, adaptive threshold output can be carried out according to the degree of overlap between the targets, a loss function can be optimized, and flexible threshold selection is provided for screening, so that the problem of missing detection caused by dense shielding of the targets is relieved, and the accuracy and recall rate of underwater group target detection are improved.

Description

technical field [0001] The invention relates to the field of image target detection, in particular to an underwater group target detection method that combines YOLOv4 and deformable convolution. Background technique [0002] In factory aquaculture, it is necessary to accurately obtain information such as the location and classification of aquaculture aquatic products for scientific aquaculture guidance such as disease early warning, growth monitoring, and bait feeding. [0003] The existing underwater group target detection method mainly uses computer vision technology, that is, the target image of the underwater aquaculture group to be detected is sent to the network model for detection, and the NMS algorithm is used for screening to obtain the location and classification information of the underwater aquaculture group. YOLOv4 is one of the commonly used convolutional neural networks. The specific network structure includes the Backbone part, the Neck part and the Predictio...

AI Technical Summary

Problems solved by technology

For this reason, image restoration, image enhancement and other methods are often used to repair images, or the network structure of enhanced feature extraction is used to extract more refined feature information, resulting in a more complex network structure.

In addition, due to the group living habits of aquatic animals, dense occlusions frequently occur in the image, and the existing YOLOv4 loss function does not have an adaptive threshold part, and only uses a fixed threshold to filter the prediction results of the NMS algorithm, which is prone to missed detection. It is difficult to guarantee the accuracy and recall rate of underwater group target detection

[0006] The deformable convolution module (DBL) can change the position of the sampling point, which can improve the modeling ability of the model, but so far there is no report on the fusion of YOLOv4 and deformable convolution to realize the detection of underwater group targets

Images