Ecs: Fresh Air Duct Leakage Detection Model for Embedded Development Environment

It is very critical to develop a new air duct leakage detection model that can run in an embedded environment to ensure the safety of the underground laboratory environment in Jinping, China. However, existing leak detection models applied in embedded development environments suffer from some problems such as long inference time and poor detection accuracy. To address these issues, this paper proposes a novel lightweight deep learning model named ECS (EfficientNet CoordConv Squeeze-and-Excitation), in which a lightweight convolutional neural network EfficientNet is leveraged to reduce the model size and CoordConv is used to improve the accuracy of leak detection. In addition, to prioritize the acquisition of computing resources for important features, an SE(Squeeze-and-Excitation) attention mechanism is added to the backbone network EfficientNet. The proposed model was trained by collecting data from the real-world fresh air duct leakage engineering scenarios, and then deployed on the embedded development board RK3588 for field testing. The experimental results show that the performance of the proposed model is superior to the five baseline models. Its size is only 15.5MB, with an average processing speed of 56.76ms per image and an average detection accuracy of 97.98%. Therefore, this work provides an effective reference for the embedded development research of fresh air duct leakage detection.