A ECA Fusion Smoke Detection Algorithm and Prototype Based on Improved YOLOv8

Currently, environmental authorities primarily regulate black smoke emissions from vehicles by sampling and testing exhaust gases using smoke meters. This regulatory method is inefficient and can pose significant health risks to testing personnel when employed for extended periods. The advancements in deep learning offer a new direction for black smoke vehicle detection; however, existing solutions have not achieved optimal results. To address issues such as excessive computational complexity and inadequate datasets, this paper employs data augmentation techniques to expand the dataset. We enhance the YOLOV8 network by incorporating depth-wise separable convolutions and attention mechanisms, presenting a lightweight object detection algorithm, YOLOV8 (ECA), deployed on edge computing devices. This algorithm achieves a detection frame rate of 20FPS on Jetson devices with a black smoke vehicle recognition rate of up to 95.57%.