Marine organism intelligent detection and identification method

Abstract

According to the marine organism intelligent detection and identification method provided by the invention, UGAN-based underwater image enhancement and depth separable convolution are introduced, so that biological category identification and detection can be carried out in various marine organisms, the difficulty of observing the marine organisms is reduced, the traditional detection mode is optimized, and the detection efficiency is improved. And high-precision and real-time marine organism detection is achieved. According to the method, morphological differences and life habits in different marine organisms and feasibility of deploying a detection algorithm in a lower memory space are considered; a lightweight deep learning network and deep separable convolution are introduced into YOLOv4, the performance and expenditure of the algorithm can be balanced, identification interference of various factors in the marine environment on marine organisms is effectively reduced, and the parameter quantity of an identification model is effectively reduced.

Description

technical field [0001] The invention relates to the technical field of deep learning and marine shellfish fry breeding, in particular to an intelligent detection and identification method based on marine organisms. Background technique [0002] With the continuous improvement of people's living standards, the market demand for precious seafood such as abalone, sea cucumber, and sea urchin is increasing day by day, which greatly promotes the vigorous development of marine fisheries represented by shallow sea aquaculture. The traditional methods of manual diving monitoring and fishing are not only time-consuming and labor-intensive, but also cause great personal injury and a high risk factor. In recent years, with the maturity and gradual application of technologies such as underwater robots, the informatization and intelligent construction of marine aquaculture has ushered in new opportunities for development. In order to realize the successful fishing of underwater robots, ...

AI Technical Summary

Problems solved by technology

[0003] Target detection algorithms have been applied in some aspects, but there are still problems in direct application to shallow sea aquaculture biometrics

The main problems in the test are as follows: 1) Due to the dim and uneven distribution of light, the clarity of the collected data is not high, and the detection accuracy cannot meet expectations; 3) Phenomena such as occlusion and half-hiding of organisms often occur, which requires higher generalization capabilities of the model

[0004] Moreover, since the embedded devices used by underwater robots, for example, often have weak computing power and small memory space, it is difficult to deploy large-scale deep learning neural networks. issues that need resolving

Images