Chemical plant safety accidents are frequent

the government has carried out a major crackdown on chemical enterprises. Compared with manual investigation with a small range and low efficiency

satellite remote sensing can efficiently monitor chemical plants in a large range

which is of great importance for production safety supervision and accident prevention in China"s chemical industry. Due to the irregular shape

strong semantic information and scattered distribution of chemical plants

the existing detection model has low precision and is difficult to be applied. In this paper

we proposed Poly-FPN object detection model

this model adopts the feature pyramid structure

improve the processing capacity of the complicated semantic target

at the same time through the proposed arbitrary quadrilateral detection head alternatives to the traditional rectangular head

realizing the objective of the irregular accurate detection. Finally

based on poly-fpn

a large area of the lower reaches of the Yangtze river was predicted

and the high-precision detection of chemical plants was realized.