Image segmentation and coverage estimation of deep-sea polymetallic nodules based on lightweight deep learning model

Area of Science:

• Marine geology and resource exploration
• Artificial intelligence and computer vision
• Mineral processing and materials science

Background:

• Deep-sea polymetallic nodules are a critical source of strategic metals, necessitating efficient exploration methods.
• Accurate assessment of nodule parameters, like Coverage Rate, is vital for evaluating deposit value.
• Existing methods for nodule data acquisition can be slow and labor-intensive, hindering effective exploration.

Purpose of the Study:

• To develop a streamlined, high-speed deep-learning model for real-time analysis of seafloor video data.
• To accurately compute the Coverage Rate parameter for polymetallic nodules.
• To enhance the efficiency and accuracy of deep-sea mineral resource exploration and evaluation.

Main Methods:

• Proposed a novel segmentation model, YOLOv7-PMN, specifically for analyzing seafloor video data.
• Replaced the YOLOv7 backbone with MobileNetV3-Small and integrated Squeeze-and-Excitation attention mechanisms.
• Employed depth-wise separable convolution modules in the head network to reduce model parameters.

Main Results:

• YOLOv7-PMN demonstrated improved detection and segmentation performance for nodules of various sizes compared to original YOLOv7.
• Achieved a 3% increase in recall rate for nodules.
• Reduced model parameters by 61.78%, memory usage by 61.15%, and increased inference speed to 65.79 FPS.
• Showcased strong generalization capabilities, reducing reliance on high-quality video data and extensive annotations.

Conclusions:

• YOLOv7-PMN is highly effective for processing complex seabed images of polymetallic nodules.
• The model offers significant improvements in speed, accuracy, and efficiency for nodule analysis.
• This technology holds substantial promise for practical application in deep-sea mineral exploration.