Fusion of crayfish optimization algorithm and MNS-YOLO for solar cell defect detection

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Construction
• Materials Science

Background:

• Solar energy integration in construction requires reliable building photovoltaics.
• Accurate solar cell defect detection is crucial for photovoltaic efficiency and reliability.
• Existing deep learning models overlook hyperparameter impacts on solar cell defect detection.

Purpose of the Study:

• To propose CMNS-YOLO, a hybrid model combining the Crawfish Optimization Algorithm (COA) with MNS-YOLO for enhanced solar cell defect detection.
• To improve feature representation and detail recovery for solar cell defects.
• To optimize network hyperparameters for superior detection accuracy.

Main Methods:

• Introduction of the C2f-MLLA module using Mamba-Like Linear Attention to enhance defect feature representation.
• Development of a Bidirectional Feature Pyramid Frequency Aware Feature Fusion network for improved detail recovery and feature fusion.
• Integration of ShapeIoU to address aspect ratio misalignment and construction of an improved MNS-YOLO network.
• Utilization of COA for MNS-YOLO network parameter optimization.

Main Results:

• The CMNS-YOLO model demonstrated improved detection accuracy by 6.3% on the PV-Multi-Defect dataset and 2.3% on the PVELAD dataset compared to the baseline model.
• The proposed model successfully maintained lightweight characteristics.
• Enhanced capability in representing target features and recovering fine details of solar cell defects.

Conclusions:

• CMNS-YOLO offers a significant advancement in solar cell defect detection accuracy and efficiency.
• The hybrid approach effectively optimizes deep learning models for photovoltaic applications.
• The method shows considerable potential for real-world deployment in solar cell inspection and maintenance.