Land desertification presents a serious challenge to global ecological security and socioeconomic development. Establishing psammophytes such as Pinus sylvestris has proven effective in reversing desertification. However, climatic and environmental pressures have caused partial withering of Pinus sylvestris, resulting in secondary soil degradation and impaired ecosystem recovery. To precisely monitor plantation status, this study introduced a feature-fusion approach combined with an improved DeepLab v3+ to address complex background interference. Colour and texture features were integrated into RGB images to enhance contrast between vegetation and surrounding objects. The improved DeepLab v3+ used MobileNetV2 as its backbone and incorporated a Step Pyramid Module (SPM), Double Branch Aggregation Module (DBAM), Convolution-Enhanced Self-Attention (CESA) and an attention gate to improve segmentation accuracy under heterogeneous conditions. Image processing was subsequently applied for further segmentation and statistical analysis of Pinus sylvestris. The segmentation results indicated that the mean values of Recall, Precision, Intersection over Union (IoU), and F1-score of the improved DeepLab v3+ improved to varying degrees compared with SwinUNet, Mask R-CNN, Segformer, HrNet_W18, UNet, and DeepLab v3+ for both the surviving and withered Pinus sylvestris. The Mean Pixel Accuracy (MPA) mean reached 87.73%. The correlation coefficients (r) between the predicted and true values for the number and area of surviving plants reached 0.90 and 0.88, respectively, while those for withered plants reached 0.92 and 0.86. These quantitative results indicated that Pinus sylvestris plantations in the Mu Us Sandy Land face severe survival challenges, which can be largely attributed to the region's harsh natural environment. Therefore, future afforestation efforts should prioritize improving survival rates through integrated management strategies to enhance the sustainability of local ecosystems.