Fish Recognition And Morphological Parameters Measurement Of Prawn Based On Computer Vision

Computer vision technology has achieved significant results in the field of object detection and recognition.At present,it has also played an important role in the field of marine aquatic products in China.Based on this technology,researchers have invented the sorting system for aquatic products,which reduce the manpower and material resources of investment.However,none of the existing researches on aquatic products could effectively measure the morphological parameters of prawn,which include the length of body,the length of carapace and the width of carapace.The related breeders can only measure manually with low efficiency.The teams that study fish recognition use foreign datasets directly.So it cannot be used for fish research in China's coastline.In this regard,we have conducted a series of experiments based on computer vision technology using shrimp and fish as research objects.And the main research will be divided into the next two contents:Firstly,in order to measure the morphological parameters of prawn,a traditional object detection method and a object detection method based on deep learning are proposed for the first time.The cascade classifier based on LBP features and the Faster RCNN based on ZF network are used to recognize the base of the eyestalk,carapace and tail of the object prawns.The length of the body,the length of the carapace and the width of the carapace are calculated by measuring the scale and the proportion of the object.The error data is obtained by comparing the measured values of the cascade classifier based on the LBP feature with the PS measurements of image.The measured values of Faster RCNN based on ZF network are compared with the measured values of PS,the measured values of the cascaded classifier based on the LBP feature and manual measurements respectively.The experimental results show that the rate of missed detection of Faster RCNN based on ZF network is lower than that of cascade classifier based on LBP features.The experimental data is more accurate,stable and efficient than the manual measurement,and the repeatability is better than it.Secondly,in order to meet the needs of fish research in China's coastline,a dataset for surveying common fishes in the southeast coast is collected for the first time,and a method based on deep learning is used.A total of 3,000 images of 7 types of 9 different fish samples were collected from different shooting angles with fins expanded or not expanded.Some of the test images were taken at night and the shooting background was different from the training images.Faster RCNN based on VGG network is used as a model to recognize fish.The experimental results show that an average classification accuracy of 96%is obtained by using VGG and Faster RCNN method,which is higher than the traditional method.And it can carry out the research work efficiently.The above two parts of the work are about intelligent identification and measurement based on computer vision technology.And it gives full play to the significance of this technology in China's coastline fish resources research.The technology not only can provide efficient shrimp breeding programs,but also can serve China's marine strategy and launch marine resource research.