Research On Fish Target Recognition Based On Acoustic Image

The assessment and monitoring of fishery resources is an important part of the protection and management of marine living resources,which is crucial for in-depth understanding of the marine fishery ecological environment.As the cornerstone of the development of marine fishery resources,fish resources are also an important biological means for the restoration of the marine ecological environment,and a favorable helper for the development of fishery economy.The extraction of fish target information is an extremely important part of fishery resources research.Accurate fish identification provides key data for the protection and management of marine fishery resources,thereby realizing the protection of marine ecosystems and the sustainable development of fishery resources.At present,the extraction of fish target information mainly relies on scientific fish finder and net capture sampling.The former mainly estimates the quantity by the reflection intensity of the fish target.The length estimation is cumbersome and the accuracy is low,and the fish biology cannot be obtained at the same time.Characteristics and positioning information,the latter method of net fishing seriously damages fish resources,consumes manpower and material resources,and its accuracy is very limited.The identification sonar is a multi-beam system that uses an acoustic lens to form a very narrow beam by focusing sound waves.It can generate acoustic images close to optical image quality in low-visibility waters,with sufficiently high image resolution and frame rate,and then the extent to which the recognition target is achieved.At present,the identification sonar in fishery research mainly stays at the stage of simple individual counting,length estimation and behavior observation,and there is no research on sonar image characteristics,detection estimation error and automatic identification.Therefore,on the basis of analyzing the principle and image features of the recognition sonar,this paper uses the recognition sonar to study the fish target recognition.The specific research contents are as follows:The imaging principle and image features of the recognition sonar are studied.This paper mainly discusses the principle of the identification sonar imaging system,analyzes the acoustic images that form the characteristics of light and dark distribution according to the echo intensity,summarizes the noise sources of the identification sonar image and the factors that affect the quality of the sonar image,and uses statistical analysis and linearity.The fitting method analyzes and verifies the fish detection length estimation from sonar images,and focuses on the length estimation error of identifying fish detected by sonar.Research ECHOVIEW’s human-computer interaction fish target recognition model.Firstly,based on the characteristics of intelligent acquisition of slant range and position information within the beam detection range by recognition sonar,ECHOVIEW’s human-computer interaction fish target recognition model is constructed,which mainly includes data integration,artificial visual calibration,sample static denoising,background removal,Trajectory tracking and target conversion to extract fish number,body length and location information.Through the fish resource detection experiment of Shanghai Chenghang Reservoir,the feasibility and accuracy of ECHOVIEW’s human-computer interaction fish target recognition model are verified.Then,the body length interval model was used to analyze the dominant range of fish targets in the reservoir area,the density estimation model was used to analyze the temporal change of fish population in the reservoir area,and the spatial distribution of fish in the reservoir area was analyzed by the interpolation spatial model,and a set of ECHOVIEW human-computer interaction fish target analysis method.An automatic fish target recognition method based on KNN background difference is proposed.Firstly,the.aris file is parsed by ARIS cope(v2.6,Sound Metric Corp,USA)and the preprocessing process of MP4 video format file is generated.Using the sonar image fish target motion characteristics,the KNN background removal and morphological expansion are used to construct a Remove speckle noise,edge detection and Deep Sort target tracking and counting algorithms,and finally achieve automatic fish target recognition.The effectiveness of the method is verified by the fixed-point pool experiment and the aerial survey experiment in the reservoir area,and the target quantity information of fish is obtained.On the basis of this method,the target pixel size feature is extracted by the target edge detection,and then the fish target length value is roughly estimated.Through the research on fish length estimation error,human-computer interaction fish recognition model,and KNN(K-Nearest Neighbor)background difference automatic fish recognition,a new method for fish target recognition in sonar images has been preliminarily formed.The target research has laid a foundation,provided a new direction for identifying sonar fish targets,filled the gap in the field of fish target research in my country,and provided technical support for the development and protection of fishery resources in my country.