Research On Precision Feeding Technology Of Crab Culture Based On Underwater Machine Vision

With the development of economy,the scale of crab culture is expanding all over the country.However,the current aquaculture management mode is still primitive and extensive,accompanied by the low efficiency of aquaculture and feed utilization rate.The efficiency and accuracy of bait feeding are two key indicators that affect the economic efficiency of river crab pond farming.Precise feeding can fully exert the efficiency of bait and obtain the maximum economic benefit.Therefore,this paper carried out research on precise feeding technology of river crab farming based on underwater machine vision.Through the preprocessing of underwater river crab images and the recognition of river crab targets,the number,size and distribution information of underwater crabs in crab ponds can be accurately obtained to solve the problem of where to feed.Through the development of self-cruising variable feeding equipment,to solve the small-area feed density distribution and range controllable efficient feeding.With the organic combination of the two parts,we can achieve efficient and accurate feed feeding in small and medium-sized crab pond culture.The main research contents of this article are as follows:A.In order to improve the imaging quality of the crab pond underwater fuzzy image,improve the image clarity and enrich the image details,thereby improving the recognition accuracy of the river crab,this paper analyzes the image based on the principle of underwater optical imaging and water absorption and scattering characteristics of light,by Histogram equalization,limited contrast adaptive histogram equalization,gamma correction,image restoration algorithm based on dark channel prior and image restoration algorithm based on image blur and light absorption.According to the principle of the first absorption of the long wave?red light?by the water body,the improved image restoration algorithm based on the dark channel prior is emphasized,and a comparative experiment is carried out to evaluate the performance of the algorithm from qualitative and quantitative analysis of image quality;B.Due to the complex underwater environment of crab ponds,large differences in river crab postures,and unclear surface texture features of river crabs in water,it has increased the difficulty of river crab recognition.Traditional target recognition methods have been difficult to meet the needs of underwater river crab recognition.In order to improve the accuracy of river crab recognition,this paper proposes a river crab recognition algorithm based on convolutional neural network,and improves the structure design of convolutional neural network.Using the improved dark channel prior image restoration algorithm based on the river crab samples to make a data set,complete the training and optimization of the improved convolutional neural network,and output the optimal river crab recognition model based on YOLO v3 improved network structure.Complete the training of river crab recognition model of Faster R-CNN with the same data set,and compare the experiment with the improved YOLO v3 river crab recognition network model to test the recognition performance;C.In order to meet the development needs of small and medium-sized crab pond aquaculture self-cruising feeding equipment,this paper studies a combined navigation system based on low-precision low-cost GPS,geomagnetic sensor and laser ranging sensor data fusion to provide multi-dimensional navigation information for feeding operation devices.In order to realize the automatic navigation operation function of the feeding operation device,this paper proposes a target path tracking strategy based on the target insertion point,and designs a course controller based on PID control;at the same time,it proposes a speed adjustment strategy for the feeding operation device straight and path switching And design a speed controller based on PID control;in order to ensure the independence of the course and speed control,a coordinated control strategy of course and speed is proposed;finally,according to the functional requirements of the feeding operation device,the software and hardware design of the feeding operation device is completed to achieve feeding operation device cruising and feeding around the standardized crab pond to solve the small-scale feed distribution density and controllable efficient feeding;The experiment through the above research shows that the restoration algorithm DCP_{{inv r,g,b}}based on the improved dark channel prior can complete the effective enhancement of the underwater river crab image.The average MSE,PSNR and Entropy indicators are 49.23,23.02and 15.73;On the premise that the quality of the underwater crab image collected is effectively guaranteed,the average class accuracy rate of the network recognition model output by the YOLO v3 crab target detection framework based on the improved convolutional neural network structure can reach 87.66%,and the crab recognition is accurate The rate is 97.74%and the recall rate is96.27%;Self-cruising feeding operation ship can effectively complete automatic navigation operation;The variable feeding device can effectively realize the functions of feeding with controllable feed distribution density and range,and the function of measuring the remaining feed in the silo;The feasibility of accurate feeding in river crab farming based on underwater machine vision can be verified.