Intelligent identification and classification of seafloor cold seep eruption states: the Haima cold seep as an example

Monitoring methane bubble plumes from seafloor cold seeps is crucial to marine research and resource exploration. Current cold seep research mainly uses remotely operated vehicles and autonomous underwater vehicles equipped with acoustic and optical devices for detection. However, issues of low accuracy remain in subsequent data analysis and image recognition, and deep learning and intelligent discrimination are urgent needs. Therefore, convolutional neural network (CNN) technology for the intelligent identification and classification of seafloor cold seep eruption states was introduced. By integrating high-quality video data obtained from marine surveys and manual identification, the bubble coverage index (BCI) was calculated. These data were used to establish a calibration dataset for cold seep eruption images based on the ResNet model. In the application of the Haima cold seep research area in the Qiongdongnan Basin, the model reached a recognition accuracy of 98.9% through comparative analysis with calibration results. This study provided a new method for data analysis and image processing and a reliable support for on-site intelligent decision-making and in-situ sample collection at seafloor cold seeps. The new method could enhance the efficiency of automated cold seep eruption image processing, solve the issue of recognition of cold seep eruption state, and offer a basis for environmental assessment and energy exploration of cold seeps.