Research On Jellyfish Detection Algorithm Based On Convolutional Neural Network

In recent years,frequent outbreaks of jellyfish in China's coastal waters have caused serious impact on human production and life,marine ecology,but the optical detection method for jellyfish is still in its infancy.Convolutional neural network is widely used in the field of target detection because of its strong feature extraction ability.Therefore,this project will use the current rapid development of deep learning theory and digital image processing theory to research the jellyfish detection algorithm based on deep learning.The main tasks are as follows:Firstly,the theoretical knowledge of convolutional neural network and transfer learning is introduced in detail,and several evaluation indicators of object detection are introduced.Secondly,it researches the production method and production process of the jellyfish detection data set,establishes a data set containing 7 species of jellyfish and fish;introduces three algorithms for underwater image preprocessing,and conducts simulation research on the three algorithms,which lays the foundation for the later detection experiment.Thirdly,in order to solve the problem that Res Ne Xt(C=32)network has a large amount of computation when it is used for jellyfish detection.The number of branches of Resne Xt(C=32)is set to 8 without significantly reducing the accuracy.Then,in order to solve the problems of some individuals too small to be detection in Resne Xt(C=8),dilation convolution is introduced into residual network.The ordinary convolution of conv4 and conv5 convolution modules in Resne Xt(C=8)is replaced with dilated convolution with expansion ratios of 2 and 4.Experiments show that the optimized multi branch expanded convolution residual network can solve the problem of low detection accuracy of small individuals in the process of jellyfish detection,but the algorithm has the problem of low real-time.Finally,to solve the problem of low real-time performance of Faster R-CNN,YOLOv3 algorithm with higher real-time performance is used to detect jellyfish.In order to improve the accuracy of YOLOv3 algorithm,the feature extraction network Darknet 53 of YOLOv3 is optimized on the basis of real-time performance.In addition,two methods are introduced in the training process.The experimental results show that the improved algorithms improve the detection accuracy of jellyfish on the premise of ensuring the detection speed,and can meet the requirements of jellyfish detection accuracy and real-time.This study lays a theoretical and technical foundation for the subsequent construction of underwater jellyfish optical imaging real-time monitoring system,and is of great significance for the development of jellyfish monitoring technology in coastal areas of China.