Discriminating marsh vegetation is critical for the rapid assessment and management of wetlands. The study area, Honghe National Nature Reserve (HNNR), a typical freshwater wetland, is located in Northeast China. This study optimized the parameters (mtry and ntrees) of an object-based random forest (RF) algorithm to improve the applicability of marsh vegetation classification. Multidimensional datasets were used as the input variables for model training, then variable selection was performed on the variables to eliminate redundancy, which improved classification efficiency and overall accuracy. Finally, the performance of a new generation of Chinese high-spatial-resolution Gaofen-1 (GF-1) and Ziyuan-3 (ZY-3) satellite images for marsh vegetation classification was evaluated using the improved object-based RF algorithm with accuracy assessment. The specific conclusions of this study are as follows: (1) Optimized object-based RF classifications consistently produced more than 70.26% overall accuracy for all scenarios of GF-1 and ZY-3 at the 95% confidence interval. The performance of ZY-3 imagery applied to marsh vegetation mapping is lower than that of GF-1 imagery due to the coarse spatial resolution. (2) Parameter optimization of the object-based RF algorithm effectively improved the stability and classification accuracy of the algorithm. After parameter adjustment, scenario 3 for GF-1 data had the highest classification accuracy of 84% (ZY-3 is 74.72%) at the 95% confidence interval. (3) The introduction of multidimensional datasets improved the overall accuracy of marsh vegetation mapping, but with many redundant variables. Using three variable selection algorithms to remove redundant variables from the multidimensional datasets effectively improved the classification efficiency and overall accuracy. The recursive feature elimination (RFE)-based variable selection algorithm had the best performance. (4) Optical spectral bands, spectral indices, mean value of green and NIR bands in textural information, DEM, TWI, compactness, max difference, and shape index are valuable variables for marsh vegetation mapping. (5) GF-1 and ZY-3 images had higher classification accuracy for forest, cropland, shrubs, and open water.

Citrus has become a pivotal industry for the rapid development of agriculture and increasing farmers’ incomes in the main production areas of southern China. Knowing how to diagnose and control citrus huanglongbing has always been a challenge for fruit farmers. To promptly recognize the diagnosis of citrus huanglongbing, a new classification model of citrus huanglongbing was established based on MobileNetV2 with a convolutional block attention module (CBAM-MobileNetV2) and transfer learning. First, the convolution features were extracted using convolution modules to capture high-level object-based information. Second, an attention module was utilized to capture interesting semantic information. Third, the convolution module and attention module were combined to fuse these two types of information. Last, a new fully connected layer and a softmax layer were established. The collected 751 citrus huanglongbing images, with sizes of 3648 × 2736, were divided into early, middle, and late leaf images with different disease degrees, and were enhanced to 6008 leaf images with sizes of 512 × 512, including 2360 early citrus huanglongbing images, 2024 middle citrus huanglongbing images, and 1624 late citrus huanglongbing images. In total, 80% and 20% of the collected citrus huanglongbing images were assigned to the training set and the test set, respectively. The effects of different transfer learning methods, different model training effects, and initial learning rates on model performance were analyzed. The results show that with the same model and initial learning rate, the transfer learning method of parameter fine tuning was obviously better than the transfer learning method of parameter freezing, and that the recognition accuracy of the test set improved by 1.02~13.6%. The recognition accuracy of the citrus huanglongbing image recognition model based on CBAM-MobileNetV2 and transfer learning was 98.75% at an initial learning rate of 0.001, and the loss value was 0.0748. The accuracy rates of the MobileNetV2, Xception, and InceptionV3 network models were 98.14%, 96.96%, and 97.55%, respectively, and the effect was not as significant as that of CBAM-MobileNetV2. Therefore, based on CBAM-MobileNetV2 and transfer learning, an image recognition model of citrus huanglongbing images with high recognition accuracy could be constructed.

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%.

Combinations of multi-sensor remote sensing images and machine learning have attracted much attention in recent years due to the spectral similarity of wetland plant canopy. However, the integration of hyperspectral and quad-polarization synthetic aperture radar (SAR) images for classifying marsh vegetation has still been faced with the challenges of using machine learning algorithms. To resolve this issue, this study proposed an approach to classifying marsh vegetation in the Honghe National Nature Reserve, northeast China, by combining backscattering coefficient and polarimetric decomposition parameters of C-band and L-band quad-polarization SAR data with hyperspectral images. We further developed an ensemble learning model by stacking Random Forest (RF), CatBoost and XGBoost algorithms for marsh vegetation mapping and evaluated its classification performance of marsh vegetation between combinations of hyperspectral and full-polarization SAR data and any of the lone sensor images. Finally, this paper explored the effect of different polarimetric decomposition methods and wavelengths of radar on classifying wetland vegetation. We found that a combination of ZH-1 hyperspectral images, C-band GF-3, and L-band ALOS-2 quad-polarization SAR images achieved the highest overall classification accuracy (93.13%), which was 5.58–9.01% higher than that only using C-band or L-band quad-polarization SAR images. This study confirmed that stacking ensemble learning provided better performance than a single machine learning model using multi-source images in most of the classification schemes, with the overall accuracy ranging from 77.02% to 92.27%. The CatBoost algorithm was capable of identifying forests and deep-water marsh vegetation. We further found that L-band ALOS-2 SAR images achieved higher classification accuracy when compared to C-band GF-3 polarimetric SAR data. ALOS-2 was more sensitive to deep-water marsh vegetation classification, while GF-3 was more sensitive to shallow-water marsh vegetation mapping. Finally, scattering model-based decomposition provided important polarimetric parameters from ALOS-2 SAR images for marsh vegetation classification, while eigenvector/eigenvalue-based and two-component decompositions produced a great contribution when using GF-3 SAR images.