| Artificial fish reefs are an important part of marine pasture construction.Due to technical limitations,the lack of relevant operating specifications,and the harsh operating environment at sea during the release process,the artificial fish reef placement position is inaccurate,which further weakened the effect of pasture construction.In this paper,acoustic methods are used to detect and identify large-scale artificial reefs on the sea floor,extract reef vector points for constraint cluster analysis and calculate the launching error,so as to provide intuitive and accurate underwater topography and geomorphology data for the launching position of artificial reefs.The main contents of this paper are as follows: using side scan sonar equipment to collect underwater topography map,utilizing digital image processing and deep learning method to automatically identify and extract artificial reef on sonar image,adopting cluster analysis method to evaluate the construction quality of the whole reef area,and then providing scientific and technical methods for the construction of marine ranch.In view of the problem that traditional visual interpretation can not solve the extraction and recognition of large-scale artificial reefs,this paper proposes an automatic recognition method of artificial reefs based on side scan sonar image.In this method,the acquired data are preprocessed by the post-processing software of the side scan sonar,and then the preprocessed image is processed by means of the mean smoothing filter,the whole denoising,the adaptive smoothing filter,the local denoising,the range filter,the sharpening image,the extraction of the edge of the target to determine the target,the binarization and so on.Finally,the reef and its acoustic image area are extracted by matrix operation,Finally,the position information of the reef is obtained.The experimental results show that the accuracy and integrity of the proposed method are over 94% and 85% respectively,and it has good generality.In addition,this paper discusses the application of fast-cnn and SSD in the detection of artificial reef targets in side scan sonar images.Aiming at the shortage of training samples,AR19 data set is constructed.Based on the data set,the Faster R-CNN model and SSD model are trained by migration learning,and verified on the test set.The experimental results show that the detection accuracy of Faster R-CNN model is 92.68%,the recall rate is 97.83%,the detection accuracy of SSD model is 51.21%,and the recall rate is 73.53%.In conclusion,deep learning algorithm can identify artificial reef accurately and efficiently in side scan sonar image,in which Faster R-CNN model has higher accuracy,robustness and generalization ability.The artificial reef can change the topography of the sea floor through the cooperation between them,and provide the habitat for the aggregation of fish.In order to evaluate the quality of reefs,it is necessary to discuss the aggregation model of artificial reefs.In this paper,the spatial clustering analysis algorithm is used to evaluate the discrete reefs and reef groups in the whole reef area.The Delaunay triangulation method is used to cluster the spatial objects,define the overall and local constraints,and summarize the quality evaluation indexes of artificial reef.Taking the data of artificial reef area in Haizhou Bay as an example,six groups of artificial reefs are obtained,which reflect the actual aggregation mode and more intuitively reflect the gap between actual placement and design.The results show that the method can reflect the dispersion degree of artificial reef clustering and the way of clustering combination,and provide quantitative value for evaluating the index error of design and measured reef points. |
PDF Full Text Request