Currently, most of the fire point algorithms based on Himawari-8/9 satellite rely on a large number of historical data modeling or spatial spectral statistical characteristics of pixels. Long term effective data are often difficult to collect and the wide applicability of algorithms is also poor. Taking advantage of the high time resolution of Himawari-8/9 satellite, and combining with the fluctuation law of brightness temperature of ground background pixels in the daily variation range of solar elevation angle, this paper proposes a new fire detection algorithm based on short-term time series window with adaptive threshold. In order to compare and verify the accuracy and universality of the algorithm, four study areas were selected globally for experimentation. The results show that the short-term time series algorithm effectively reduces the dependence on historical data, has wide applicability, and reduces the false alarm fire point caused by the poor universality of the empirical threshold and other factors to a certain extent. The accuracy of fire point recognition is significantly improved.

In this paper, we focus on measuring the similarity of sentence pair. Noting that a single sentence vector may lose fine-grained semantic information which is important for sentence matching, we propose an embedding matrix to calculate a multi-granularity similarity matrix and find the true semantic alignment of two sentences. We also propose a semantic importance calculation and semantic decomposition that are simple but effective. The proposed model does not require any sparse features or external resources such as WordNet. Compared with other state-of-the-art models, we successfully train in a short time and achieve competitive results on similarity measurement and paraphrase identification tasks. Experiments and visual analysis show the good performance and interpretability of the model.