Extraction of Nearshore Aquaculture Raft Regions through Spatial-frequency Domain Feature Fusion Network

The aim of this study is to develop a method based on spatial frequency domain feature fusion network to efficiently and accurately extract the nearshore aquaculture raft region from remote sensing images. Due to the problems of insufficient accuracy and low degree of automation in traditional methods for the identification of aquaculture rafts under complex hydrological and meteorological conditions, spatial frequency analysis and deep learning techniques were introduced in this work, and a new framework for raft region extraction combining these two technologies was proposed. Firstly, the spatial frequency domain of optical remote sensing images is analyzed to obtain the boundary and texture features of aquaculture rafts. By combining traditional image processing techniques with frequency analysis, the differences between the subtle structures in the raft area and the adjacent natural water bodies are highlighted. These feature spectra are used as input to the deep learning model to realize the localization and feature extraction of the initial raft region. Further, a feature fusion network is developed by combining residual network(ResNet). This network carries out multi-level fusion processing for frequency features at different scales, and extracts richer and more effective feature information to identify the raft region by deepening the network structure layer by layer. Using a large number of labeled remote sensing image data to train the network, the generalization ability and robustness of the model are enhanced. The test results show that compared with traditional methods, the feature fusion network proposed in this study has significantly improved in various performance indicators, especially in terms of recognition accuracy and robustness under complex hydrometeorological conditions. This technology can realize automatic and efficient extraction of nearshore aquaculture raft area on remote sensing images at different times and in different regions, which has important application value for monitoring and management of aquaculture resources. In summary, the results of this study not only improve the accuracy and efficiency of remote sensing image interpretation in coastal aquaculture raft area, but also provide a new environmental monitoring and management tool for aquaculture industry.