A Semantic Segmentation Algorithm for Water Plants Distribution Detection
Download as PDF
DOI: 10.23977/CNCI2020086
Author(s)
Wangyi Wang, Jiaoyan Ai, Jianwu Zheng and Dongyi Yao
Corresponding Author
Jiaoyan Ai
ABSTRACT
The planting density of water plants in landscape lake is very important to the water quality restoration. In order to accurately divide the area without water plants, this paper proposes a novel deep learning algorithm to explain the distribution of water plants, and the method based on the improved FC - DenseNet network structure, its structure is mainly consists of five encoding networks and the five decoding networks. The encoding network and the decoding network are connected together by concat layer, the last layer of the network is a pixel-level classifier. The encoding network structure is improved on the basis of DenseNet network, and the decoding network restores the image to the original resolution through corresponding up-sampling, so as to obtain a more accurate segmentation effect. Compared with FCN and other seven semantic segmentation network models, the results show that this research method has better segmentation performance and can accurately segment the sediment area without water plants.
KEYWORDS
Dataset; density detection; semantic segmentation; deep learning