The 4th place in the preliminary of National Digital Ecological Innovation Contest - Intelligent Algorithm Contest (2021)
Date:
Zifan Chen, Jie Zhao, Li Zhang
This competition is based on high-resolution remote sensing image data of different landforms. Its target is to use intelligent remote sensing image interpretation technology to identify and extract land cover and use types. We propose a dual-path segmentation structure based on UNet++, which uses both semantic and structural information of the image and establishes connections between multiple resolutions. Bagging integration of five models under five-fold cross-validation helps us win fourth place in the preliminary round. In the rematch, the complexity of the model needs to be considered. We prune the redundant connections with high resolution to improve the performance further. However, we failed to obtain a good ranking in the second round, limited by my knowledge and technical level.
