资源环境科技发展态势分析平台

Ammonia (NH₃) is a precursor of atmospheric aerosols. The lack of real-time updated and high spatial resolution atmospheric NH₃ point source information is one of the critical problems in the study of atmospheric pollution. Satellite-based NH₃ observation data provide a new approach to NH₃ point source detection. However current spatial resolution of NH₃ concentration satellite data is still not sufficient to extract NH₃ point sources at regional scale with high accuracy. Based on the traditional oversampling method to improve the spatial resolution of NH₃ satellite data, this paper used high-resolution satellite image data with the geometric morphological features of NH₃ point sources to detect the NH₃ point sources at regional scale under deep learning techniques. The high-precision NH₃ point source location and category information at regional scale is extracted concurrently, solving the current problem that multiple point sources cannot be detected in a single satellite NH₃ concentration hotspot. After analyzing a priori knowledge, the industrial sources were further subdivided. The atmospheric ammonia point source detection technique in this paper were used in Northeast China, where the number and types of NH₃ emission sources are abundant. The detection results significantly increased the number of NH₃ potential point sources detected in Northeast China compared with the traditional oversampling method. The deep learning-based technology combines NH₃ concentration remote sensing products with high resolution satellite image data to effectively detect the spatial distribution and types of NH₃ point sources at regional scale, providing powerful technical support for NH₃ emission regulation and control.