Research On Identification Method Of Coral,Coral Reef Fish And Starfish Based On Deep Learning

Coral reef ecosystem has very important ecological functions.It plays an extremely important role in both marine life and human beings.Organisms such as corals,coral reef fishes and starfish are important components of coral reef communities,and their growth,population size and living habits can reflect the state of the coral reef ecological environment,they have high research value.Therefore,it is of great significance to do research on the detection and classification of them.At present,there are problems in the detection and classification of marine organisms,such as few available datasets and low recognition accuracy.In recent years,with the rapid development of deep learning,breakthroughs have been made in many fields such as computer vision.The research methods of object detection and classification based on deep learning technology have been paid more and more attention in practical applications.It has become the main method in the field of marine biology that collecting underwater coral data through advanced underwater imaging systems and then using deep learning methods to analyze them.In this paper,YOLO v5,the object detection algorithm,is applied to the detection and classification of corals,coral reef fish and starfish,which provides strong support for the analysis and research of coral reef ecosystems.The main research work of this paper is as follows:First of all,the paper used web crawler technology based on Python to crawl images of corals,coral reef fishes and starfishes,then deduplicate and clean the obtained images to construct their respective image datasets.The final coral dataset contains 34 genera,12792 images.The final coral reef fish dataset contains 72 families,23061 images.The final starfish dataset contains 37 families,2428 images.Secondly,the paper propose an improved YOLO v5 s algorithm that introduces the SEnet attention mechanism,and construct a new target detection dataset based on three self-built datasets and some publicly available marine life images.Training and testing the improved algorithm on this dataset,the overall m AP50 reached 93.6%,which is 2.4% higher than the original algorithm,which shows that the performance of the improved model has improved.Thirdly,based on several classic convolutional neural network models and the YOLO v5x-cls pre-training model with SEnet,the paper carries on the image classification experiments on the self-built coral,coral reef fish and starfish datasets.The highest top5 accuracy of the improved YOLO v5x-cls on these three datasets is 74%,88.5% and 79.2%,respectively.Compared with the CNN model,the improved YOLO v5x-cls has better overall performance.Finally,based on the above experiments and the trained models,the paper build a system for detecting and classifying marine organisms,which can detect the input image,determine species,then classify it to obtain specific categories.This paper constructs a dataset of three marine organisms including coral,coral reef fish and starfish.In addition,the paper propose an improved YOLO v5 algorithm that introduces the SEnet attention mechanism,and the improved algorithm is used in object detection and image classification on the self-built dataset.This research method provides a new idea for marine biology-related research,and the research results have certain reference significance.