Near-shore seabed fish detection and tracking statistical method based on neural network

Abstract

The invention discloses a near-shore seabed fish detection and tracking statistical method based on a neural network. The method comprises the following steps: FcycleGAN image migration processing is carried out on an input underwater real-time video to generate clear images; then, features of fishes in the video are extracted by inputting the images into a basic neural network Darknet53 for processing, wherein the features mainly comprise shape features, texture features, and the like, of the fishes; a detection branch is divided into two detection stages to finally output specific positions and types of the fishes, and a tracking branch outputs a Jacobian matrix and a distance vector of fish mass point swimming; then fishes in a certain range of a predicted position are matched with fishes in a previous position. Therefore, positions, types, and serial numbers of fishes in each image are obtained.

Description

technical field [0001] The invention relates to the field of seabed exploration and detection, in particular to a neural network-based detection and tracking statistics method for near-shore seabed fish. Background technique [0002] The ocean has very rich biological resources; therefore, coastal countries are vigorously developing marine ranching, especially fishery-increasing breeding-type marine ranching. The Food and Agriculture Organization of the United Nations, the Food and Agriculture Organization of the United Nations, recorded global food fish production from marine ranches at 28.7 million tons ($67.4 billion) in 2016, accounting for 49.5% of total world aquaculture production in 2016. Overexploitation, aquaculture is also in a saturated state; therefore, marine ranching management is considered to be an important way to solve the decline of fishery resources, however, there are also some problems in marine ranching management (such as overfishing, ecosystem imbal...

AI Technical Summary

Problems solved by technology

Finally, a hybrid neural network model was constructed, however, the main limitation of this approach is that it treats fish as particles and cannot classify fish types, Marini et al. (2018) developed a genetic programming-based content-based image analysis methods, however, when a large number of fish gather in front of the camera, the crowded scene limits the recognition efficiency, and when these gatherings are particularly dense, individuals often overlap each other, which increases the false negative rate

The current existing technology is aimed at the classification and recognition of underwater fish images. The problem with them is that this type of method is not conducive to the statistics of fish schools, and will cause serious errors. To distinguish whether it is the same fish as the previously identified fish, so there will be redundant errors when counting the number

Images