Research on adaptive ocean remote sensing target detection framework: An efficient solution based on the broad learning system

With the rapid advancement of ocean satellite remote sensing technology and the growing availability of ocean observation data, the demand for efficient and highly adaptable target detection techniques has become increasingly urgent. To address this challenge, this study introduced an adaptive ocean remote sensing target detection framework based on the Broad Learning System (BLS), characterized by its shallow architecture and rapid incremental learning capabilities. The framework, named Auto-Features-BLS (AF-BLS), automatically and efficiently selects the optimal combination of pretrained feature extractors from diverse machine learning models via the Hyperopt library and processes them using BLS to detect ocean targets. The AF-BLS model was evaluated on 12 types of ocean targets, including internal waves, ships, rain cells, and ocean fronts. Experimental results demonstrate that AF-BLS exhibits strong robustness and flexibility in detecting these targets, outperforming traditional models with an average Accuracy of 99.19%, an average Precision of 98.51%, an average Recall of 98.15%, an average F1-Scores of 98.33%, and an average Matthews Correlation Coefficient (MCC) of 96.33% on the testing set. Furthermore, the AF-BLS model trained on CPU significantly improves overall efficiency, with training speed nearly five times faster and inference speed more than three times faster than conventional GPU-based models, highlighting its practicality for deployment in resource-constrained or real-time scenarios. Additionally, the study used internal waves as an example to validate the model’s generalization performance across untrained sensors and global applications. The proposed AF-BLS model offers an efficient and highly adaptable solution for ocean remote sensing target detection.