Deep learning techniques have boosted the performance of hyperspectral image (HSI) classification. In particular, convolutional neural networks (CNNs) have shown superior performance to that of the conventional machine learning algorithms. Recently, a novel type of neural networks called capsule networks (CapsNets) was presented to improve the most advanced CNNs. In this paper, we present a modified two-layer CapsNet with limited training samples for HSI classification, which is inspired by the comparability and simplicity of the shallower deep learning models. The presented CapsNet is trained using two real HSI datasets, i.e., the PaviaU (PU) and SalinasA datasets, representing complex and simple datasets, respectively, and which are used to investigate the robustness or representation of every model or classifier. In addition, a comparable paradigm of network architecture design has been proposed for the comparison of CNN and CapsNet. Experiments demonstrate that CapsNet shows better accuracy and convergence behavior for the complex data than the state-of-the-art CNN. For CapsNet using the PU dataset, the Kappa coefficient, overall accuracy, and average accuracy are 0.9456, 95.90%, and 96.27%, respectively, compared to the corresponding values yielded by CNN of 0.9345, 95.11%, and 95.63%. Moreover, we observed that CapsNet has much higher confidence for the predicted probabilities. Subsequently, this finding was analyzed and discussed with probability maps and uncertainty analysis. In terms of the existing literature, CapsNet provides promising results and explicit merits in comparison with CNN and two baseline classifiers, i.e., random forests (RFs) and support vector machines (SVMs).

Postclassification comparison (PCC) and change vector analysis (CVA) have been widely used for land use/cover change detection using remotely sensed data. However, PCC suffers from error cumulation stemmed from an individual image classification error, while a strict requirement of radiometric consistency in remotely sensed data is a bottleneck of CVA. This letter proposes a new method named CVA in posterior probability space (CVAPS), which analyzes the posterior probability by using CVA. The CVAPS approach was applied and validated by a case study of land cover change detection in Shunyi District, Beijing, China, based on multitemporal Landsat Thematic Mapper data. Accuracies of “change/no-change” detection and “from-to” types of change were assessed. The results show that error cumulation in PCC was reduced in CVAPS. Furthermore, the main drawbacks in CVA were also alleviated effectively by using CVAPS. Therefore, CVAPS is potentially useful in land use/cover change detection.

Accurate mapping of winter wheat over a large area is of great significance for guiding policy formulation related to food security, farmland management, and the international food trade. Due to the complex phenological features of winter wheat, the cloud contamination in time-series imagery, and the influence of the soil/snow background on vegetation indices, there remains no effective method for mapping winter wheat at a medium spatial resolution (10–30 m). In this study, we proposed a novel method called phenology-time weighted dynamic time warping (PT-DTW) for identifying winter wheat based on Sentinel 2A/B time-series data. The main advantages of PT-DTW include (1) the use of phenological features in two periods, i.e., the greenness increase before winter and greenness decrease after heading, which are common to all winter wheat and are distinct from the features of other land cover types, and (2) the use of the normalized differential phenology index (NDPI) instead of traditional vegetation indices to provide more robust vegetation information and to suppress the adverse impacts of soil and snow cover, especially during the before-winter growth period. The proposed PT-DTW method was employed for winter wheat mapping based on Sentinel 2A/B data on the Huang-Huai Plain, China. Validation with visually interpreted samples showed that the produced winter wheat map achieved an overall classification accuracy of 89.98% and a kappa coefficient of 0.7978, outperforming previous winter wheat classification methods. Moreover, the planting area derived from PT-DTW agreed well with census data at the municipal level, with a coefficient of determination of 0.8638, indicating that the winter wheat map produced at 20 m resolution was reliable overall. Therefore, the PT-DTW method is recommended for winter wheat mapping over large areas.

In the past decades, spectral unmixing has been studied for deriving the fractions of spectrally pure materials in a mixed pixel. However, limited attention has been given to the collinearity problem in spectral mixture analysis. In this paper, quantitative analysis and detailed simulations are provided, which show that the high correlation between the endmembers, including the virtual endmembers introduced in a nonlinear model, has a strong impact on unmixing errors through inflating the Gaussian noise. While distinctive spectra with low correlations are often selected as true endmembers, the virtual endmembers formed by their product terms can be highly correlated. It is found that a virtual-endmember-based nonlinear model generally suffers more from collinearity problems compared to linear models and may not perform as expected when the Gaussian noise is high, despite its higher modeling power. Experiments were conducted on a set of in situ measured data, and the results show that the linear mixture model performs better in 61.5% of the cases.

As a nondestructive geophysical tool, Ground penetrating radar (GPR) has been applied in tree root study in recent years. With increasing amounts of GPR data collected for roots, it is imperative to develop an efficient automatic recognition of roots in GPR images. However, few works have been completed on this topic because of the complexity in root recognition problem. Based on GPR datasets from both controlled and in situ experiments, the randomized Hough transform (RHT) algorithm was evaluated in root object recognition for different center frequencies (400 MHz, 900 MHz, and 2000 MHz) in this paper. Reasonable accuracy was obtained (both a high recognition rate and a low false alarm rate) in these datasets, which shows it is feasible to apply the RHT algorithm for root recognition. Furthermore, we evaluated the influence of root and soil factors on the recognition. We found that the performance of RHT algorithm is mainly affected by root interval length, root orientation, and clutter noise of soil. The recognition results by RHT could be applied for large scale root system distribution study in belowground ecology. Further studies should be conducted to reduce clutter noise and improve the recognition of the complex root reflections.