Research On The Vitality Analysis Of Shrimp Fry Based On Computer Vision

In recent years,with the rapid development of China’s aquaculture industry,fine intelligence as well as preventive farming methods have gradually become the future trend of aquaculture.As one of the aquaculture species,multidimensional factors such as rapid temperature changes,abnormal dissolved oxygen or human intervention during aquaculture put shrimp fry in a state of stress and affect their survival rate.Timely detection of shrimp fry in a state of stress can significantly reduce farming losses and improve shrimp farming productivity and efficiency.General acute stress causes abnormal physiological and behavioral changes in shrimp fry,so shrimp behavior is a valuable reference information for farmers.In this paper,we establish a method to quantify the vigor of shrimp fry based on practical experience in the production process,and use it to investigate the behavioral changes of shrimp fry under stress caused by man-made external factors,using South American white shrimp as the research object.The main research contents and results are as follows.(1)Pre-processing study of shrimp fry stress images: a video acquisition system was established to obtain video data of shrimp fry behavior before and after stress,and then median filtering and noise removal were performed on shrimp fry images,and histogram equalization was performed on the images to enhance the contrast of shrimp fry against a blurred background,while grayscale images of shrimp fry were extracted to avoid distortion of visible bands.(2)Study of shrimp fry stress behavior feature extraction method: firstly,the shrimp target is segmented,the shrimp skeleton is extracted using morphological processing to achieve the measurement of the relative length of the shrimp,and the regularity of the shrimp fry is determined by the length of the shrimp;in order to initially determine the changes in the movement of shrimp fry before and after stress,the Lucas-Kanade optical flow method is used to obtain the direction vector of shrimp movement,calculate and analyze the direction vector distribution of shrimp fry Then,the color changes,surface ripple jitter,projection changes and changes in the trajectory and behavior of shrimp before and after stress were quantified by selecting two parameters,H and S,as color features in HSV color space,and extracting four covariates,namely energy,entropy,moment of inertia and correlation,as texture features using the gray scale coevolution matrix(GLCM).Finally,the shrimp fry are tracked by Deep SORT,and the shrimp trajectory is described by combining the improved weighted average method with the average swimming speed and average swimming acceleration as the motion feature parameters to finally quantify the vitality of the shrimp fry.The combination of kinematic parameters to characterize the vigor of shrimp fry reduces the misclassification rate and improves the detection accuracy.(3)Research on shrimp fry vigor detection and identification: a shrimp fry vigor identification model based on XGBoost+RFE was established,and a total of 12 features were selected for modeling.Firstly,the features were input to RFE,and the importance ranking of shrimp fry stress behavior feature parameters was repeatedly screened until the ranking no longer changed,and then the features were input to XGBoost for re-scoring,and the feature ranking was derived through multiple iterations,and the features were combined to select the feature groups F12,F5,F7 and F1 with the highest accuracy and real-time uniformity,and each feature of this feature group was assigned different The features of this feature group were given different weights,and repeated experiments were conducted to fuse the features and finally form the shrimp fry vigor detection model.The results show that when the weights of evaluation factors F12,F5,F7 and F1 are taken as 0.30,0.26,0.25 and 0.19,respectively,the fused features have the highest recognition accuracy of 98.96%,which is 6.98%,2.4% and 1.96% higher than the previous methods of single color,single texture and combination of optical flow and texture,respectively,while this model is more accurate than BP,KNN and SVM detection accuracies by 1.31%,3.69% and 2.13%,respectively.The study showed that the method not only has strong robustness but also shortens the detection time,which provides a research basis for the analysis of shrimp fry vigor intensity and is important for the selection of shrimp fry as well as culture monitoring.