A Novel Semantic Segmentation Approach Using Improved SegNet and DSC in Remote Sensing Images

Introduction

As a vital component of the earth's ecosystem, vegetation serves an irreplaceable role in climate regulation, maintaining environmental viability and ensuring the earth's ecological balance. As countries strengthen the implementation of sustainable development policies, vegetation coverage is gradually becoming a key focus area. The distribution and accurate identification and classification of vegetation are the premises and foundations for studying vegetation coverage.

The automatic analysis of remote sensing (RS) data is a prerequisite for mining information and transforming RS observations into knowledge (Alsmirat et al., 2019; Kumbhojkar & Menon, 2022; Wang et al., 2020). Its main purpose is to establish a unified, compact, and semantic representation of large RS datasets, thereby laying the foundation for subsequent information mining. The automatic analysis of RS datasets mainly includes data expression, retrieval, and understanding (Kumar et al., 2022; Lv et al., 2022; Stergiou et al., 2021). At present, RS is mostly used to obtain the dynamic information and images of vegetation in real time, but the images obtained are largely affected by external factors such as the local geographical environments, so it is difficult to realize accurate classification (Kadri et al., 2022; Kotaridis & Lazaridou, 2021; Kumbhojkar & Menon, 2022; Wenjuan & Shao, 2021). Consequently, the problem of vegetation classification in RS images is a research hotspot (Chopra et al., 2022; Yu et al., 2020).