Research On Feature Recognition Of Tuna Purse Seine Fish School In The Western And Central Pacific Ocean Based On Deep Learning

As Tuna fishery,as the "gold industry of ocean fishery",has long been concerned by the country.The tuna Seine fishery of mainland China developed rapidly in the first 13 years through the continuous resource expansion and production scale of the state.In 2013,the Chinese mainland had 14 tuna seining boats in the central and western Pacific,catching 81,000 tons of tuna,accounting for 5.1 percent of the total production in the region.But in its rapid development,encounter some problems such as the modernization level is not high,fishing boats technology research needs to be strengthened,plus international tuna prices continue to decline,the fishing cost calculated on work days to make tuna purse Seine boat unit price rose sharply,China and the west Pacific fishery management organization tuna resources at present,the methods for the management of the Seine ships operation implement stricter,etc.,lead to the Chinese mainland China and the west Pacific Ocean tuna purse Seine,fishing production is declining in 2018 fell to 14000 tons.The tuna Seine fishery in the central and western Pacific is facing severe challenges.How to improve the production mode of fishing boats and how to enhance the high efficiency of tuna Seine fishing has become the key consideration and concern of China's Seine fishery production and management.Through the study of fishing for tuna purse Seine fishery situation shows that FADs after being extended,China tuna purse Seine fleet average search time of fish increased dramatically,indicating that the Seine fleet are not familiar with the changes of time and space of free fish distribution regularity,so that the boat can't quick look and found that the fish,if you can't find the fish,Seine assignments will not be able to undertake,can only be "wandering around in the ocean,the cost of consumption.Traditional fishing methods,such as using binoculars for extended periods of time,looking for tuna from high towers,and even renting manned helicopters to inspect tuna stocks at high prices,are time-consuming and costly.In order to improve the efficiency of searching for tuna fish and reduce the cost of searching for fish,it is of great significance to research on fish target identification based on deep learning for tuna Seine fishery.The recognition technology can be run on uav onboard computer image recognition technology to replace the traditional fish search method,help to faster and more efficient to find free fish,fishing boats to reduce fishing boats sailing waiting time,avoid the waste of time,blind search fishing grounds,improve the utilization rate of work days,eventually find fish consumed by reducing the number of days and reduce the fuel cost,improve the tuna purse Seine fishing efficiency of our country and international competitiveness,make the enterprise get better profits and further improve the country's fisheries status.In order to achieve fast classification and recognition of tuna fish characteristics,this paper analyzes the large Shanghai started fishing co.,LTD in China and the west Pacific Ocean tuna fish of tuna purse Seine homework during the collection of video data,analyzing the characteristics of the behavior of tuna fish,cut and screened for shoal characteristics identification training images,making a tuna feature image data sets,build deep learning model to complete the identification of tuna fish.The main work is to study the fish feature recognition in several aspects such as image acquisition,feature labeling and simulation recognition,and to verify its performance in fish feature recognition by experiments.The main work contents and research results are as follows:(1)Analysis of tuna purse Seine in China mainland in China and the west Pacific historical changes in production,analysis of Chinese and western Pacific bluefin tuna fisheries management policy change,show the policy changes on the influence of tuna purse Seine fleet production in our country,studies have shown that FADs after being extended,a dramatic increase of the average time for fish catches in our country,this shows that the tuna purse Seine fleet in China mainland for free fish search efficiency is very low and need to strengthen and improve search method for free fish.(2)In this paper,a large number of tuna Seine data and video images of tuna fish are collected,and the key points of the collection process are discussed.Before collecting fish characteristics,it is necessary to analyze the visual characteristics of tuna schools at sea with a large number of tuna sea videos,and collect them in combination with actual tuna scenes.In order to improve the effect of image recognition,it is necessary to remove the pictures unrelated to the features,so as to ensure that each picture has the features to be extracted and the features are clearly visible.(3)Based on the collected feature pictures,feature labeling is carried out.Feature tagging refers to the artificial identification and analysis of the features of fish image.The feature information of the image is selected in the box to judge the characteristics of the feature information.In the tagging process,fish swarm features have various irregular shape targets.In order to detect performance,the tagging method of minimum target boundary box should be adopted to minimize other influencing factors in the selected area,so as to improve the later identification accuracy.(4)An application model based on deep learning for fish identification in Seine nets of tunas in the central and western Pacific was implemented.,According to the research of the tuna fish identification features,from the perspectives of model training and select the target recognition algorithm,and conform to the current study by fish target detection framework to test the shoal video,will contain the fish image input to the shoal characteristics detection module,the output characteristics of fish detection module is used to identify the fish of the full image,get the fish identification results,images and video to identify and design methods.The experimental results show that it is feasible to search tuna stocks by using image recognition technology.