Study On Evaluation Method Of Artificial Reef Construction Based On Machine Vision

In recent years,faced with the problems of declining resources and pollution of the marine environment,artificial reef construction and research have been increasingly valued by coastal countries.In the artificial reef construction project,planning,designing,and launching work are the beginning of the construction of artificial reefs,and the supervision and benefit evaluation of artificial reefs is a long-term task.The appropriateness of monitoring and assessment methods directly affects the effectiveness of artificial reefs.The construction of artificial reefs is developing in the direction of information and intelligence,massive images captured by its video surveillance system need to be handled in a more intelligent way.In this paper,combined with machine vision technology,the fish identification and tracking methods are mainly studied.This research will help to achieve the continuity and intelligence of artificial reef construction evaluation,and at the same time,it can also improve the design and layout of artificial reefs and provide technical support for the construction of smart seas in China.This paper conducts research work through theoretical analysis and experimental verification.The main contents and results of the study are as follows:(1)Taking four kinds of goldfish as the research object,in the laboratory environment,the collection of images and videos of four kinds of goldfish is completed.In order to achieve the recognition of the fish body,a corresponding sample set was established in this paper.In the sample set,150 images of each goldfish were used,for a total of 600 images.Completion of preprocessing of images in the sample set,including image cropping,image graying,image size standardization,etc.(2)Research on Theory and Methods of Image Processing.After reading a large number of documents and books,according to the actual needs of the goldfish image processing,the pretreatment of fish images is completed in the MATLAB software.The research mainly focuses on image graying,enhancement processing,spatial domain filtering,segmentation,and image edge detection.The principle of the image is also discussed.Specific examples are used to operate the image,is selected to process the image collection.The images make very important preparations for the follow-up image recognition and tracking through these pre-processing methods.(3)Before the identification of fish,the characteristics of the fish need to be extracted.In this paper,principal component analysis(PCA)and speed-up robust features(SURF)are used to extract features of fish images.This paper proposes a method based on Hu invariant moments and SURF features to obtain fusion features,which improves the robustness of image features.(4)This paper uses BP neural network to classify four different types of goldfish.Using genetic algorithm to optimize BP neural network and improve the recognition rate of the algorithm.This paper compares the effects of PCA features and Hu-SURF fusion features on image recognition rate,and finds that the combination of feature fusion and genetic algorithm optimization neural network combination method has a high recognition rate,and the total recognition rate reaches 86%.(5)In order to realize the detection and tracking in a complex background,Gaussian mixture model and Kalman filter are used to achieve the detection and motion estimation of the moving target.The target can be effectively detected and drawn with the function.The trajectory of the fish provides technical support for the study of fish body behavior.When tracking specific fish,the traditional SURF-based method is prone to out-of-focus phenomenon.Based on this,a SURF-KLT method is proposed to realize real-time stable tracking of specific fish.