Existing perception and identification methods for coal–rock interfaces are generally based on various cutting signals during shearer mining, but they cannot achieve pre-perception. Further, some pre-identification methods are significantly affected by the mining environment and cannot achieve accurate identification. In this study, a universal method is proposed to achieve pre-perception and accurate recognition of coal–rock interfaces. Using the coal–rock interface identification test-bed, active excitation infrared images of coal–rock were tested with different excitation time, distance, and intensities, and an image dataset was built for training the universal network model. This was done primarily to improve the universality of the model for infrared image identification of coal–rock under various parameter conditions. Second, the pyramid pooling module and MobileNetV2 were used to effectively extract the semantic features from the infrared images. Meanwhile, a convolutional block attention module was employed to improve the coal–rock interface identification ability of the proposed network. Finally, the efficiency of the proposed network model was tested on the infrared image dataset. The experimental results demonstrated that the memory occupied by the proposed network model is 9.12 MB, the test time is 38.46 ms/piece, and the intersection of union of coal and rock is 98.07 and 98.38%, respectively. Additionally, the pixel accuracy of coal and rock is 98.68 and 99.50%, respectively, which is significantly higher compared to other network models. Based on the constructed multi-parameter universality dataset, the proposed identification model of a coal–rock interface has good adaptability to active excitation infrared images acquired by different parameters and could provide the theoretical foundation and technical premise to achieve automated and intelligent mining.

To accurately describe the features of a remote sensing image by considering the relationship in the neighbourhood system, this paper presents a neighbour mapping and manifold projection-based image segmentation algorithm called NM_MP (Neighbour Mapping and Manifold Projection, NM_MP). First, the features of the image are described by Gaussian distributions. Then, the image described by the Gaussian distributions is mapped to a manifold called the Riemannian manifold that is able to characterize the patterns of objects in remote sensing images. To fully utilize the advantages of the expression ability of the Riemannian manifold space, a data submanifold and a parameter submanifold are established to depict the features of the image and the corresponding segmentation result. Through projecting points from the data submanifold onto the nearest candidate on the parameter submanifold and updating the candidates according to the projection results, the candidates tend to approach optimum segmentation. The NM_MP algorithm is validated on synthetic and real remote sensing images. The experimental analysis demonstrates that the NM_MP algorithm can effectively decrease the impact of noise and outliers and consequently obtain promising segmentation results on remote sensing images.

Stable and continuous remote sensing land-cover mapping is important for agriculture, ecosystems, and land management. Convolutional neural networks (CNNs) are promising methods for achieving this goal. However, the large number of high-quality training samples required to train a CNN is difficult to acquire. In practice, imbalanced and noisy labels originating from existing land-cover maps can be used as alternatives. Experiments have shown that the inconsistency in the training samples has a significant impact on the performance of the CNN. To overcome this drawback, a method is proposed to inject highly consistent information into the network, to learn general and transferable features to alleviate the impact of imperfect training samples. Spectral indices are important features that can provide consistent information. These indices can be fused with CNN feature maps which utilize information entropy to choose the most appropriate CNN layer, to compensate for the inconsistency caused by the imbalanced, noisy labels. The proposed transferable CNN, tested with imbalanced and noisy labels for inter-regional Landsat time-series, not only is superior in terms of accuracy for land-cover mapping but also demonstrates excellent transferability between regions in both time series and cross-regional Landsat image classification.

Scholars have identified various factors for adopting green IT by organizations from different perspectives. However, the studies about the priority of those factors and the interdependent relations among them are relatively lack. To address this problem, we use the Motivation-Opportunity-Ability (MOA) framework to establish a systematic model of the factors which drive firms’ green IT practices. Furthermore, an integrated Decision-Making Trial and Evaluation Laboratory (DEMATEL) and Analytic Network Process (ANP) approach has been adopted to analyze the complex relations among different factors and determine the weight distribution of them. The results show that top managers supports, policy and regulation, institutional pressure and human resources are key factors driving firms’ green IT practices. This study provides the implications for organizations and policy makers to understand and develop strategies to implement green IT.

The cascade of convolution layers and the end-to-end training process facilitate CNN feature extraction and transmission, and promote the success of CNN in image processing. However, the drawback of heavily relying on large-scale high-quality training samples restricts its applications. To avoid costly and unrealistic manual annotations for large-scale remote sensing images, existing land cover maps are considered as an alternative to manual annotations, in which noisy labels are inevitable. To alleviate the impact of noisy labels, this paper proposes to improve the consistency feature learning ability of CNNs as a feasible solution in practical land cover mapping. Firstly, an intraclass feature consistency constraint is introduced to maintain the consistency of CNN feature maps for the same class. Then, an inter-iteration feature consistency constraint is employed to guide the network to learn features that are consistent with the whole underlying distribution inside a mini-batch. These two feature consistency constraints work in a cooperative and complementary manner with the traditional cross-entropy, and together improve the consistency feature learning ability of the proposed Feature Consistency Network (FCNet). Experimental results demonstrate the effectiveness of the proposed FCNet. Extensive experiments on different network structures validate the generalization of the proposed feature consistency constraints.