Under the open light path condition, the spectral characteristics of polluted gases and atmospheric components are overlapped, making it difficult to directly identify the polluted gases. This study proposes an adaptive feature extraction method, which pre-generates the spectral features under various atmospheric conditions. The rapid feature extraction is performed using the Lasso algorithm for selecting the optimal target-background combination, reconstructing the background spectrum, and extracting the target features. The effectiveness of the proposed algorithm is verified via the methane remote detection under different backgrounds; the ammonia gas detection is also performed under different relative humidity conditions along with the indoor close-range ethylene detection. The proposed method is compared with the Harig's method. The results show that the proposed method can well eliminate background and possesses strong practicability.