Investigating the spatial patterns of fishing vessel operations with a dual-driven recognition-spatial analysis framework

Accurate recognition of fishing vessel operation types is crucial for enhancing fishery management performance, yet existing methods lack in-depth exploration of spatial behavioral patterns. This study proposes a dual-driven recognition-spatial analysis framework. Methodologically, the framework integrates a Multi-scale Convolutional Neural Network (MSCNN), Bidirectional Long Short-Term Memory (BiLSTM), and dual-attention mechanisms to construct a deep learning recognition model, with hyperparameters automatically optimized via Particle Swarm Optimization (PSO). Subsequently, a spatial analysis module is developed by combining Kernel Density Estimation (KDE), Local Indicators of Spatial Association (LISA), and Standard Deviational Ellipse (SDE) to identify operational hotspots, reveal spatial patterns, and quantify monthly migration trends. Unlike existing approaches that treat recognition and spatial analysis as separate tasks, this framework integrates them into a recognition-driven spatial analysis pipeline, enabling accurate identification and systematic quantification of spatiotemporal patterns in a unified manner. Empirically, the model achieves a recognition accuracy of 95.47% on a real-world vessel trajectory dataset. Further spatial analysis reveals that purse seine operations exhibit highly clustered stability, gillnet operations display significant seasonal variations, and trawl operations maintain a southeast-northwest axial expansion pattern. This research offers a novel approach to the interpretation of fishing vessel behavioral patterns and the development of intelligent fishery management systems.