Research On Image Recognition Of Sea Creatures Based On Python

The harvest of benthic organisms such as sea cucumbers and sea urchins is mainly carried out manually.These sea creatures grow in a complex environment,and the huge water pressure on the seabed during harvest can seriously damage the health of the fisherman.At present,more and more automated fishing equipment has begun to enter this field,but the detection and positioning of marine organisms has always been the technical bottleneck of automated fishing.In this thesis,the real-time underwater image acquisition,target detection and scale measurement of sea cucumbers,sea urchins and other sea creatures are deeply studied.In this thesis,sea cucumber,sea urchin,starfish and scallop images obtained through field acquisition and network search for the research object.A real-time image acquisition system of underwater marine life is built first.The system has the characteristics of light weight and strong expansibility in structure,and can complete high-definition video shooting with the assistance of artificial light source in real-time acquisition.Secondly,using digital image processing and deep learning technology,the research on the detection algorithm of marine organisms is carried out.A marine life image database was established,and the existing images were augmented by offline data augmentation.In order to solve the problems that traditional target detection algorithms have low recognition rate and are easy to miss detection in complex underwater environments,the improved Faster-RCNN algorithm is used in marine creature target detection.This algorithm uses a residual structure to map the input to the output layer.This method avoids the problem of gradient disappearance or network degradation caused by the increase of the number of network layers and enhances the feature extraction ability of the model through feature fusion,improves the accuracy of target detection.The algorithm is compared with the Res Net-SSD algorithm.The comparison experiment shows that the recognition accuracy of the algorithm is 18.4% higher than that of the Res Net-SSD algorithm based on meeting the requirements of real-time image processing.Thirdly,this thesis proposes a marine biological scale measurement and depth calculation method based on parallel structured light.The method uses parallel line lasers to generate active structural information on the bottom of the water and uses the mapping relationship between pixel distance and spatial distance to realize rapid real-time calculation of marine organism scale and depth.Based on the single CCD image of Marine organisms,species identification and scale and depth measurement are achieved.Finally,A real-time measurement system for marine organisms based on CCD and parallel structured light is built,and uses the Tensor Flow framework to embed the software into lightweight computer equipment,and conduct pool experiments.Experiments show that: within the sampling time of 45 ms,the identification of marine species with an accuracy of 89.7% is achieved;the scale identification error of marine creatures using parallel structured light is within ±7%.A basis for species identification,scale and depth measurement methods and measurement devices for automated fishing of marine organisms is provided by this study,which can be effectively applied to automated fishing equipment for sea cucumbers,sea urchins,starfish and scallops.