Cavity Convolution and Attention Mechanism Based Smoke Anomaly Detection Model for 10kV Distribution Room

The distribution room is a critical facility for electrical equipment. However, its operation and inspection currently rely on manual periodic checks, which are insufficient for handling unexpected situations. To address this limitation, a smoke anomaly recognition model for 10kV intelligent distribution rooms is proposed, utilizing cavity convolution and an attention mechanism. The model is based on the Yolov8 target detection algorithm. In the feature extraction network, a cavity convolution module is incorporated to expand the network's receptive field by introducing gaps between kernel elements. This enhancement enables the network to capture features over a wider range, thereby improving smoke detection accuracy. Additionally, the integration of spatial and channel attention mechanisms allows the model to allocate higher attention to both the target-specific areas and corresponding channels during feature extraction. Furthermore, the model replaces the traditional anchor frame-based detection approach with an adaptive head detector. The adaptive head dynamically generates anchor frames based on the input shape, accommodating various input resolutions and aspect ratios. This adaptability enhances the model's ability to detect smoke anomalies of different sizes and scales. Experimental results demonstrate that the proposed method achieves an average recognition accuracy of 95% and a recall rate of 94%, validating its effectiveness in detecting smoke anomalies in distribution rooms.