Mangrove-forest classification by using deep learning algorithms has attracted increasing attention but remains challenging. The current studies on the transfer classification of mangrove communities between different regions and different sensors are especially still unclear. To fill the research gap, this study developed a new deep-learning algorithm (encoder–decoder with mixed depth-wise convolution and cascade upsampling, MCCUNet) by modifying the encoder and decoder sections of the DeepLabV3+ algorithm and presented three transfer-learning strategies, namely frozen transfer learning (F-TL), fine-tuned transfer learning (Ft-TL), and sensor-and-phase transfer learning (SaP-TL), to classify mangrove communities by using the MCCUNet algorithm and high-resolution UAV multispectral images. This study combined the deep-learning algorithms with recursive feature elimination and principal component analysis (RFE–PCA), using a high-dimensional dataset to map and classify mangrove communities, and evaluated their classification performance. The results of this study showed the following: (1) The MCCUNet algorithm outperformed the original DeepLabV3+ algorithm for classifying mangrove communities, achieving the highest overall classification accuracy (OA), i.e., 97.24%, in all scenarios. (2) The RFE–PCA dimension reduction improved the classification performance of deep-learning algorithms. The OA of mangrove species from using the MCCUNet algorithm was improved by 7.27% after adding dimension-reduced texture features and vegetation indices. (3) The Ft-TL strategy enabled the algorithm to achieve better classification accuracy and stability than the F-TL strategy. The highest improvement in the F1–score of Spartina alterniflora was 19.56%, using the MCCUNet algorithm with the Ft-TL strategy. (4) The SaP-TL strategy produced better transfer-learning classifications of mangrove communities between images of different phases and sensors. The highest improvement in the F1–score of Aegiceras corniculatum was 19.85%, using the MCCUNet algorithm with the SaP-TL strategy. (5) All three transfer-learning strategies achieved high accuracy in classifying mangrove communities, with the mean F1–score of 84.37%~95.25%.

Combining machine learning algorithms with multi-temporal remote sensing data for fine classification of wetland vegetation has received wide attention from researchers. However, wetland vegetation has different physiological characteristics and phenological information in different growth periods, so it is worth exploring how to use different growth period characteristics to achieve fine classification of vegetation communities. To resolve these issues, we developed an ensemble learning model by stacking Random Forest (RF), CatBoost, and XGBoost algorithms for karst wetland vegetation community mapping and evaluated its classification performance using three growth periods of UAV images. We constructed six classification scenarios to quantitatively evaluate the effects of combining multi-growth periods UAV images on identifying vegetation communities in the Huixian Karst Wetland of International Importance. Finally, we clarified the influence and contribution of different feature bands on vegetation communities’ classification from local and global perspectives based on the SHAP (Shapley Additive explanations) method. The results indicated that (1) the overall accuracies of the four algorithms ranged from 82.03% to 93.37%, and the classification performance was Stacking > CatBoost > RF > XGBoost in order. (2) The Stacking algorithm significantly improved the classification results of vegetation communities, especially Huakolasa, Reed-Imperate, Linden-Camphora, and Cephalanthus tetrandrus-Paliurus ramosissimus. Stacking had better classification performance and generalization ability than the other three machine learning algorithms. (3) Our study confirmed that the combination of spring, summer, and autumn growth periods of UAV images produced the highest classification accuracy (OA, 93.37%). In three growth periods, summer-based UAVs achieved the highest classification accuracy (OA, 85.94%), followed by spring (OA, 85.32%) and autumn (OA, 84.47%) growth period images. (4) The interpretation of black-box stacking model outputs found that vegetation indexes and texture features provided more significant contributions to classifying karst wetland vegetation communities than the original spectral bands, geometry features, and position features. The vegetation indexes (COM and NGBDI) and texture features (Homogeneity and Standard Deviation) were very sensitive when distinguishing Bermudagrass, Bamboo, and Linden-Camphora. These research findings provide a scientific basis for the protection, restoration, and sustainable development of karst wetlands.

No abstract