Operating Mechanism Detection in Aluminum Electrolysis Workshops via YOLOv8-MIE

Accurate identification and tracking of crane operational tools in aluminum electrolysis plants are crucial for the intelligent transformation of smelting infrastructure. However, the field faces significant challenges, such as continuous scale variations, target ambiguity in complex environments, and uneven lighting conditions, all of which collectively diminish the precision and robustness of existing algorithms. To address these issues, we introduce a YOLOv8 variant specifically designed for aluminum electrolysis, termed YOLOv8-MiE. This model’s backbone incorporates an enhanced MobileNetV3, integrated with Convolutional Block Attention Modules (CBAM), surpassing traditional squeeze-and-excitation methods to strengthen the extraction of key features and enrich their representation, thus enhancing the model’s discriminative power. The neck structure of YOLOv8-MiE features a C2f-iRMB-EMA setup, combining inverted residual mobile blocks with advanced multi-scale attention mechanisms. This integration reduces computational costs while enabling the model to effectively handle various scale features in the visually demanding electrolysis environment. To further optimize feature exploitation, YOLOV8-MiE employs Adaptive Spatial Feature Fusion (ASFF) strategies, dynamically integrating multi-scale features based on real-time detection requisites. Empirical validations attest to the efficacy of our approach, manifesting in a 3.2% enhancement in mean Average Precision (mAP), a 6.4% boost in accuracy, and a conspicuous improvement in computational efficiency, tailored specifically to a dataset designed for aluminum electrolysis crane operations.