Research On Object Detection Technology Of Sea Creatures Based On Deep Learning

With the increase in the demand for underwater seafood,studying the object detection of underwater marine organisms can improve the problem of high risk factors for artificial fishing operations,short working hours,and major physical harm.The robot can replace people to complete the sea cucumber fishing task and can also be used later.Facilitate the development and utilization of marine resources.Sea cucumber is currently a key sea-bottom organism that is caught on the seabed.Sea cucumber is the object of study and can represent most of the seabed organisms.Therefore,the object of underwater object detection in this study is sea cucumber,and the sea cucumber is classified and three-dimensionally positioned.In order to solve the problems of traditional submarine underwater environment fishing products,such as the poor accuracy of traditional underwater object classification and the low efficiency of three-dimensional positioning time,this paper proposes a submarine biological target detection method based on deep learning.In this paper,underwater image enhancement is first performed,then the object is classified and positioned based on deep learning,and the regression box is used as a binocular localization of the region of interest to improve the accuracy of the object classification and the efficiency of positioning.The main research work is as follows:This article first briefly introduces the development process of deep learning.Then,it thoroughly analyzes the research status and existing problems of underwater target detection technology.Based on these issues,the object detection method based on deep learning that this paper intends to adopt is also given.Secondly,the sea cucumber data set was established,and the image expansion method was given for the problem of scarcity of sea cucumber image data;for underwater light absorption and scattering effects,problems such as fog and color cast in underwater images were improved,improved image enhancement techniques for white balance dark channel priority principles.Thirdly,the structure elements of traditional convolutional neural networks and the classical neural network structure are analyzed.Then the background detection methods based on convolutional neural network are used to detect the background misdetection and missed detection of underwater sea cucumbers.The object detection the network and can accurately classify and locate the coordinates of the sea cucumber's two-dimensional regression box.Finally,the study of underwater binocular three-dimensional positioning,aiming at underwater imaging quality is far less than the water problem,through the improved image enhancement technology of this paper,the number of binocular feature points extraction is improved,and the accuracy of positioning detection is improved;Due to the long matching time,the regression box obtained through deep learning is used as a stereo feature matching for the region of interest,which reduces the matching search range and improves the speed of location detection.In this paper,multiple sets of experiments are used to verify the feasibility and effectiveness of the deep learning target detection method for improving target classification and improving three-dimensional positioning efficiency.At the same time,relevant structures and types of the network are improved according to specific problems.Finally,this paper summarizes and prospects the practical effects and prospects of target detection based on deep learning,and points out the existing deficiencies and problems to be further explored.