Research On Accurate Determination Of Feeding Quantityfor River Crab Farming Based On Machine Vision

Since bait expenses account for about 70% of the cost of river crab farming,scientifically determining the amount of bait is a key technical link in the process of river crab farming.At this stage,river crab feeding mainly relies on the farmers’ farming experience to estimate the total amount of bait in the pond on the day,and then the bait is scattered in the pond through the method of one person supporting the boat and one person feeding.However,this method is not based on the principle of distribution according to needs,and it is easy to cause repeated bait throwing or missing throwing in the pond area.After bait throwing,it is only possible to judge whether the amount of bait is appropriate by manually punting the pond to observe the food counter.It can be seen that this method relies heavily on manual experience and extensive operations,which can easily lead to low bait utilization,poor quality of river crabs,and poor breeding benefits.In view of the above problems,a machine vision-based method for accurately determining the feeding amount of river crabs was proposed.The specific content is as follows:(1)Machine vision-based algorithm for river crab and bait detectionIn order to realize the detection of river crabs and bait targets,a detection algorithm for river crabs and bait targets based on improved YOLOv5 s was proposed to solve the problems of complex underwater environment,large target size difference and slow detection speed.YOLOv5 s model was improved by Ghost structure,Bi FPN structure and CA attention mechanism.And the improved model was compared with other target detection models for analysis.Furthermore,the test results showed that the m AP of the improved model in this paper was 96.6%,and the calculation volume was 8.5GFLOPs.Compared with the original YOLOv5 s model,Its average precision improved by 1.2 percentage points,and the calculation volume and model memory were reduced by 40%.Above,on the Android phone,the average detection speed of the improved model was 146ms/frame,which was 17.78% faster than the original model,and maintained a good detection effect,as well as balanced the model detection accuracy and performance requirements for speed.(2)Studied on the target location method of river crabTo explore the distribution of crabs in the pond,it is necessary not only to detect the number of crabs,but also to collect the location information of the crabs.Therefore,the positioning technology was studied.Firstly,different positioning coordinate systems and the conversion between them were studied.Then the INS/GNSS tight combined navigation and positioning method based on extended Kalman filter was adopted.Finally,the effectiveness of the positioning method was verified through the experiment of fixed point positioning of feeding boats.Furthermore,the experimental results showed that that the average error of INS/GNSS integrated navigation and positioning after extended Kalman filtering is 1.3 cm.Compared with the GNSS positioning without data fusion,the maximum error was reduced by 68.4%,which could meet the positioning accuracy requirements of this paper.(3)Builded an information management system for river crab farmingIn order to make the data involved in the crab farming process more conveniently managed,the river crab farming information management system was built.Additionally,the system detection part included feeding boat and electronic food counter,where the feeding boat was used to explore the distribution of river crabs and provide a decision-making basis for determining the baiting amount,and the electronic food counter was used to investigate the remaining bait and the water quality in the sub-region of the pond and provide feedback for adjusting the baiting amount.Meanwhile,the management platform was built based on the Java programming language,and functions such as data transmission,data storage and data display were realized,and the river crab farming information management website was developed to provide a reference solution for the implementation of aquatic product networking.In conclusion,firstly the distribution of river crabs was collected by the proposed machine vision detection algorithm and positioning method.Secondly,the distribution map of river crab feeding density combined with the total amount of baiting for the day was generated.Thirdly,the baiting task command was sent to the feeding boat,and the boat performed the baiting operation according to the task.Finally,the remaining bait was detected by the machine vision algorithm to provide quantitative feedback on the remaining baiting amount for the next accurate baiting operation.In the meantime,the experimental results showed that intelligent detection,real-time positioning and intelligent early warning had been realized,and each sub-area of the pond was fed on demand according to the feeding density map,which improved the utilization rate of bait and increased the river crab farming efficiency.