YOLOv5 marine fish target detection method based on attention mechanism and DropBlock

Abstract

The invention discloses a YOLOv5 marine fish target detection method based on an attention mechanism and DropBlock. The YOLOv5 marine fish target detection method comprises the following steps: step 1, obtaining a target fish data set; step 2, labeling and dividing the obtained data set; step 3, constructing a YOLOv5 marine fish target detection model added with an attention mechanism and DropBlock; and 4, performing fish detection by adopting the constructed marine fish target detection model. The objective of the invention is to solve the problems that according to an existing fish detection method, due to the fact that the activity range of marine fishes is large and various living environments exist, the fishes shot underwater deform, clear fish targets cannot be captured, the fish detection accuracy is affected, and the fish detection efficiency is affected. And extraction of fish features by a single convolutional neural network is susceptible to interference of image noise, and fishes cannot be rapidly detected, so that the speed and precision of a target detection method are affected.

Description

technical field [0001] The invention belongs to the technical field of fish research, in particular to a YOLOv5 marine fish target detection method based on an attention mechanism and DropBlock. Background technique [0002] There are many species of marine fish and a wide range of activities. Relying on manual statistics is inefficient and difficult. In addition, most of the fish targets are photographed underwater, and there are water turbidity, fluctuations, and light disturbances, resulting in low definition of the fish targets in the pictures. [0003] In the prior art, the patent document with the application publication number CN112926652A discloses a deep-learning-based fish fine-grained image recognition method, which selects the public marine fish video data set, which cannot be well adapted and extracted from the sea area of ​​the country When using the convolutional neural network fused with candidate regions to detect fish images, the detection speed is not fas...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to solve the problem that in the existing fish detection method, due to the large range of activities of marine fish and various living environments, the fish photographed underwater are deformed and cannot capture relatively clear fish targets, which affects the quality of the fish. However, the fish features extracted by a single convolutional neural network are easily disturbed by image noise, and cannot detect fish quickly, thus affecting the speed and accuracy of the target detection method.

Images