Fish Feeding Behavior Detection Method Under Circulating Aquaculture Based On Computer Vision

China is the largest aquaculture country in the world.Studying the technical methods of aquaculture and improving the aquaculture efficiency are of great significance to the aquaculture industry in China.In aquaculture,fish feeding techniques and feeding methods are critical.This is because in the feeding process,underfeeding or overfeeding will affect the economic benefits of fish farming.However,in the current fish farming,the breeding density gradually increases,and the traditional manual feeding methods are increasingly not suitable for modernized fine culture.In the artificial feeding process,the feeding amount is easily affected by the subjective experience and subjective consciousness of the farmer,and it is difficult to accurately grasp the feeding amount and it is impossible to accurately feed.With the continuous development of farming methods and scale,modern aquaculture urgently needs a new type of efficient feeding method.Accurate feed control of fish has become a research hotspot in recent years.Studying the feeding rules of fish schools,formulating reasonable feeding strategies,improving the feeding methods of fish schools,and improving the utilization rate of fish feeds are important ways to improve the economic benefits of aquaculture in China.Among them,studying the feeding patterns of fish schools is the basis for achieving all other goals.The method of manual observation has been difficult to apply in modern large-scale farming.With the development of computer vision technology,it has become an efficient and reasonable method to use computer vision technology to detect and observe the feeding law of fish school.Taking Cyprinus carpio speculari as an experimental object,this paper employs computer vision technology to detect feeding behavior by using shape and texture information of fish in the process of feeding.Firstly,the image is subtracted,grayed out,binarized and so on,and the image shape and texture information are obtained.Then,the feeding behavior of fish is classified by BP neural network.The main work of this paper is divided into the following two parts:1.Capture the video sequence before and after feeding the fish and cut the video to extract the pictures before and after feeding the fish.Then,perform background subtraction,graying,and binarization on these pictures to extract the shape and texture of the target picture.information.This method uses the characteristic information of the entire fish school image as input to avoid the behavioral tracking of single fish.Compared with the previous methods,this paper takes the interference factors such as splash and ripple as part of the image texture information to avoid the adverse effects caused by these factors and further simplifies the computational complexity.2.This paper not only considers the texture information of the image,but also further combines the shape features of the image.Using the two features together as the input,the BP neural network is used to detect the feeding behavior of the fish school.In this paper,three characteristic quantities of entropy,deficit moment and shape parameters of fish school images are extracted.The input state of BP neural network is used as the input value of the BP neural network by using the moment of inequality,entropy,shape parameters and entropy,respectively,to detect and identify the fish feeding state.The results show that compared with the single detection method based on texture features,the accuracy rate of this method is 98%.Compared with the former,the proposed method has better fitting,smaller value and better performance.Changes in fish feeding behavior were detected.Although the use of a single texture feature will result in a shorter running time,in practice,such a small time unit is not required to be very accurate,so this method can be better applied to the detection of fish feeding status.