Fish multi-target tracking method based on balanced joint network

Abstract

The invention discloses a fish multi-target tracking method based on a balanced joint network, and belongs to the technical field of aquaculture. According to the method, on the basis of a Chinese agricultural artificial intelligence innovation and entrepreneurship contest data set, optimization, arrangement and supplementation are carried out on the basis of an original data set, and a new OptMFT data set is generated; the robustness of the model in a complex environment is further enhanced by performing merging training on a data set; meanwhile, some negative sample image frames with poor image quality are removed, the training weight of data with obvious fish targets and clear swimming tracks is enhanced, the fish recognition precision of the model is further improved, and full-frame training is performed on a data set to enhance robustness verification of the model under the condition of high-speed swimming of fish schools; the method is wide in application range, can achieve a good effect in multiple breeding environments, and is high in practicability.

Description

technical field [0001] The invention belongs to the technical field of aquaculture, in particular to a fish multi-target tracking method based on a balanced joint network. Background technique [0002] (1) Defects and problems of aquaculture in traditional intensive farming ponds [0003] For the aquaculture model in intensive aquaculture ponds, the level of dissolved oxygen in the water will have an important impact on the feeding of fish, the digestion of bait, and the growth of fish. When the dissolved oxygen content is too low, it will cause fish to have adverse symptoms such as decreased appetite, low digestion and absorption rate, and growth retardation. From the perspective of fish physiology and behavior, low dissolved oxygen will lead to fish clusters, floating heads, diving behavior, and even rapid and irregular swimming on the water surface, dark and light body color, and white film in fish eyes. Therefore, the multi-target tracking method of fish can be used to...

AI Technical Summary

Problems solved by technology

[0010] (1) Problems with underwater image data: the fish multi-target tracking data set mainly has two problems: data acquisition and data frame quality

[0013] (2) Problems with traditional tracking methods

Since 2013, although the target tracking method based on deep learning has made a series of significant progress, because the actual scene is often more complex than the evaluation data, and when the target detection algorithm based on deep learning is migrated to the fish target, it is limited. The impact of the underwater environment has a high demand for image enhancement, so the current tracking algorithm cannot meet the needs of robustness, real-time and accuracy at the same time

[0017] (3) Fish occlusion problem

[0020] Second, schools of fish often experience sudden acceleration when feeding or when they are startled, and the explosive acceleration can cause the target to be lost

By recording high frame rate video data, the problem of target loss caused by acceleration can be effectively solved, but this method will significantly increase the burden on hardware devices

In model training, the increase of image frames will also lead to a decrease in training efficiency

Images