Ammonia Gas Detection Based on CNN with Heatmap and Transfer Learning

Abstract

In this paper, a new gas detection method based on artificial intelligence was proposed. First, the sensory data of ammonia gas is converted into a heatmap, and the changing state of ammonia gas concentration is analyzed with the improved neural network of transfer learning. Second, from the qualified candidate heatmaps, the rising turning point that represents the ammonia adsorption sensing can be found. Based on the above process, the state of ammonia leakage will be clearly presented. By accurately detecting and finding the rising turning point, the leak ammonia concentration can be known, and then a gas prediction system can be established. The dataset used in this study comes from the sensing data obtained by the ammonia sensor in this study under different operational temperatures and different ammonia concentrations condition. Experimentally, the best operating temperature of the ammonia sensor in this study is 190°C. At this temperature, the sensitivity of 1000 ppm NH3/air reaches 15239%. In addition, the lowest detecting concentration of ammonia is 20 ppb NH3/air. This studied ammonia sensor has the advantages of small size, low cost, wide sensing concentration (20 ppb~1000 ppm NH3/air) and wide operating temperature (25~225°C), and super high ammonia sensitivity. Summary, the studied sensor has excellent ammonia sensing characteristics, and can perform detection and prediction in real time.