In hyperspectral image classification, a large number of labeled samples are necessary for deep network training. However, labeling of hyperspectral images is tedious, difficult, and time-consuming work. In this letter, a new active learning (AL) framework for deep networks is proposed. An auxiliary deep network for the basic learner is constructed to learn the uncertainty of unlabeled samples in the candidate data set. Both the features of the original training data and the features of the middle hidden layer of the basic learner are gathered into a fully connected network with a newly defined loss function. In order to avoid the problem of an insufficient number of samples or a large amount of computation in AL, data sampling is performed on a candidate data set, and augmentation is conducted on newly selected samples. The proposed model is evaluated on different data sets and compared with other related methods. The results show that the proposed model performs better on different data sets than other methods.

With the rapid development of earth-observation technology, the amount of remote sensing data has increased exponentially, and traditional relational databases cannot satisfy the requirements of managing large-scale remote sensing data. To address this problem, this paper undertakes intensive research of the NoSQL (Not Only SQL) data management model, especially the MongoDB database, and proposes a new approach to managing large-scale remote sensing data. Firstly, based on the sharding technology of MongoDB, a distributed cluster architecture was designed and established for massive remote sensing data. Secondly, for the convenience in the unified management of remote sensing data, an archiving model was constructed, and remote sensing data, including structured metadata and unstructured image data, were stored in the above cluster separately, with the metadata stored in the form of a document, and image data stored with the GridFS mechanism. Finally, by designing different shard strategies and comparing MongoDB cluster with a typical relational database, several groups of experiments were conducted to verify the storage performance and access performance of the cluster. The experimental results show that the proposed method can overcome the deficiencies of traditional methods, as well as scale out the database, which is more suitable for managing massive remote sensing data and can provide technical support for the management of massive remote sensing data.

Band Selection is a research hotspot in the field of hyperspectral imaging (HSI) processing. This paper proposes a method that selects bands for HSI classification by the explainability of a convolutional neural network (CNN). We design a CNN architecture and use its 1D gradient-weighted class activation mapping (GradCAM) to obtain a gradient-weighted heatmap (GradHM) by the last layer in the well-trained CNN. Since the pooling layer in the CNN leads to a dimension change of the GradHM, cubic spline interpolation (CSI) is used to up-sample the GradHM. To further improve the accuracy of the up-sampled GradHM for the important band, guided backpropagation was adopted to obtain a more detailed GradHM (Guided-GradHM). Finally, based on GradHM and Guided-GradHM, two strategies, named Average Selection (AS) and Total Selection (TS), are proposed to form new and different combinations of methods (GradHM+AS, GradHM+TS, Guided-GradHM+AS, and Guided-GradHM+TS). In experiments, the proposed methods show better performance compared to other methods. In most cases, if the selected bands were fewer, then the Guided-GradHM+AS was more credible than the other methods. Furthermore, different datasets were used to validate the proposed methods.

In most of the existing studies on the band selection using the convolutional neural networks (CNNs), there is no exact explanation of how feature learning helps to find the important bands. In this letter, a CNN-based band selection method is presented, and the process of feature tracing is explained in detail. First, a 1-D CNN model is designed and trained to reach high accuracy. Next, the derivative of the sum of partial output combinations of a layer is obtained with respect to the input layer of the CNN. Then, the derivative maps are used to obtain the sample saliency maps and class saliency maps (CSMs). Finally, the bands are selected using the CSMs. The proposed model is verified by experiments on different data sets and compared with other related methods. The results show that the proposed model can achieve better performance than the other methods.

The development of industrial infrastructure in the Beijing–Tianjin–Hebei(BTH) region has been accompanied by a disorderly expansion of industrial zones and other inappropriate development. Accurate industrial heat source classification data become important to evaluate the policies of industrial restructuring and air quality improvement. In this study, a new classification of industrial heat source objects model based on active fire point density segmentation and spatial topological correlation analysis in the BTH Region was proposed. First, industrial heat source objects were detected with an active fire point density segmentation method using NPP-VIIRS active fire/hotspot data. Then, industrial heat source objects were classified into five categories based on a spatial topological correlation analysis method using POI data. Then, identification and classification results were manually validated based on Google Earth imagery. Finally, we evaluated the factors influencing the number of industrial heat sources based on an OLS regression model. A total of 493 industrial heat source objects were identified in this study with an identification accuracy of 96.14%(474/493). Compared with results for nighttime fires, the number of industrial heat source objects that were identified was higher, and the spatial coverage was greater; the minimum size of the detected objects was also smaller. Based on the function of the identified industrial heat source objects, the objects in the BTH region were then divided into five categories: cement plants (21.73%), steel plants (53.80%), coal and chemical industry (12.66%), oil and gas developments (7.81%), and other (4.01%). An analysis of their operations showed that the number of industrial heat source objects in operation in the BTH region tended to first rise and then decline during the 2012–2021 period, with the peak being reached in 2013. The results of this study will aid the rationalization of industrial infrastructure in the BTH region and, by extension, in China as a whole.

The surge of deep learning has catalyzed considerable progress in self-supervised Hyperspectral Anomaly Detection (HAD). The core premise for self-supervised HAD is that anomalous pixels are inherently more challenging to reconstruct, resulting in larger errors compared to the background. However, owing to the powerful nonlinear fitting capabilities of neural networks, self-supervised models often suffer from the Identity Mapping Problem (IMP). The IMP manifests as a tendency for the model to overfit to the entire image, particularly with increasing network complexity or prolonged training iterations. Consequently, the whole image can be precisely reconstructed, and even the anomalous pixels exhibit imperceptible errors, making them difficult to detect. Despite the proposal of several models aimed at addressing the IMP-related issues, a unified descriptive framework and validation of solutions for IMP remain lacking. In this paper, we conduct an in-depth exploration to IMP, and summarize a unified framework that describes IMP from the perspective of network optimization, which encompasses three aspects: perturbation, reconstruction, and regularization. Correspondingly, we introduce three solutions: superpixel pooling and uppooling for perturbation, error-adaptive convolution for reconstruction, and online background pixel mining for regularization. With extensive experiments being conducted to validate the effectiveness, it is hoped that our work will provide valuable insights and inspire further research for self-supervised HAD. Code: https://github.com/yc-cui/Super-AD.