Steel Smoke Detection Method Based on Improved AFPN with Attention Mechanism

Detecting by-product smokes from processes such as sintering and coking in the steel production process is crucial for efficient and safe steel manufacturing. However, existing detection methods suffer from inefficiency, complex computations, and low accuracy. To solve these problems, a steel smoke detection method based on improved AFPN with attention mechanism is proposed in this paper. The proposed method employs FasterNet as the backbone to eliminate redundant feature extraction inherent in standard convolution, thereby significantly improving feature extraction speed. Secondly, Tiny AFPN integrates the attention mechanism with Partial Convolution (PConv), which is proposed to enhance the efficiency and precision of feature fusion. Additionally, a multi-head detector based on Spatial Attention Mechanisms (SAM) is designed to improve detection accuracy for targets of varying sizes. In order to verify the efficiency and accuracy of the proposed method, the actual steel smoke samplers are used. The experimental results show that the proposed method has significant advantages in both detection speed and accuracy compared with traditional models.