Landcover classification is an important application in remote sensing, but it is always a challenge to distinguish different features with similar characteristics or large-scale differences. Some deep learning networks, such as UperNet, PSPNet, and DANet, use pyramid pooling and attention mechanisms to improve their abilities in multi-scale features extraction. However, due to the neglect of low-level features contained in the underlying network and the information differences between feature maps, it is difficult to identify small-scale objects. Thus, we propose a novel image segmentation network, named HFENet, for mining multi-level semantic information. Like the UperNet, HFENet adopts a top-down horizontal connection architecture while includes two improved modules, the HFE and the MFF. According to the characteristics of different levels of semantic information, HFE module reconstructs the feature extraction part by introducing an attention mechanism and pyramid pooling module to fully mine semantic information. With the help of a channel attention mechanism, MFF module up-samples and re-weights the feature maps to fuse them and enhance the expression ability of multi-scale features. Ablation studies and comparative experiments between HFENet and seven state-of-the-art models (U-Net, DeepLabv3+, PSPNet, FCN, UperNet, DANet and SegNet) are conducted with a self-labeled GF-2 remote sensing image dataset (MZData) and two open datasets landcover.ai and WHU building dataset. The results show that HFENet on three datasets with six evaluation metrics (mIoU, FWIoU, PA, mP, mRecall and mF1) are better than the other models and the mIoU is improved 7.41–10.60% on MZData, 1.17–11.57% on WHU building dataset and 0.93–4.31% on landcover.ai. HFENet can perform better in the task of refining the semantic segmentation of remote sensing images.

Rice is China’s main crop and its output accounts for 30% of the world’s total annual rice production. Rice growth status is closely related to chlorophyll content (called Soil and Plant Analyzer Development (SPAD) values). The determination of a SPAD value is of great significance to the health status of rice, agricultural irrigation and regulated fertilization. The traditional SPAD value measurement method is not only time-consuming, laborious and expensive but also causes irreparable damage to vegetation. The main aim of the present study is to obtain a SPAD value through the inversion of hyperspectral remote sensing images. In order to achieve this purpose, the hyperspectral image of rice at different growth stages at the canopy scale was first acquired using a hyperspectral imaging instrument equipped with a drone; the spectral characteristics of the rice canopy at different growth stages were analyzed and combined with a ground-level measured SPAD value, the bands with high correlation between the SPAD values and the spectra of the rice canopy at different fertility stages were selected. Subsequently, we combined the spectral characteristics with the continuous projection algorithm to extract the characteristic band and used the PLS method in MATLAB software to analyze and calculate the weight of each type of spectral value and the corresponding canopy SPAD value; we then used the wavelength corresponding to the spectral value with the highest weight as the used band. Secondly, the four methods of univariate regression, partial least squares (PLS) regression, support vector machine (SVM) regression and back propagation (BP) neural network regression are integrated to establish the estimation model of the SPAD value of rice canopy. Finally, the models are used to map the SPAD values of the rice canopy. Research shows that the model with the highest decision coefficient among the four booting stage models is “booting stage-SVR” (R² = 0.6258), and the model with the highest decision coefficient among the four dairy maturity models is “milk-ripe stage-BP” (R² = 0.6716), all of which can meet the requirement of accurately retrieving the SPAD value of rice canopy. The above results can provide a technical reference for the accurate, rapid and non-destructive monitoring of chlorophyll content in rice leaves and provide a core band selection basis for large-scale hyperspectral remote sensing monitoring of rice.

Recognition and classification of road-related landslides are a critical requirement in pre- and post-disaster hazard analysis. They are primarily done through field mapping or manual image interpretation from commercial satellites images. This paper developed a Normalized Difference Road Landside Index (NDRLI)-based method to delineate road-induced landslides and enhance their presence in remotely sensed digital imagery based on free Landsat Operational Land Imager (OLI) sources. The NDRLI-based method includes NDRLI, Shape Index of Spectral Curve (SISC), and other optimizing steps such as deleting shadow and slope <20° area to recognise landslides. The test results show that the NDRLI-based method is effective in extracting road-induced landslide information, although the Kappa coefficient should be further improved.