The efficient classification of remote sensing images (RSIs) has become the key of remote sensing application. To tackle the high computational cost in the traditional classification method, in this paper we propose a new RSI classification method based on improved convolutional neural network (CNN) and support vector machine (SVM) (CNN-SVM). In this method, we first designed a seven-layer CNN structure and took the ReLU function as the activation function. We then inputted the RSI into the CNN model and extracted feature maps and replaced the output layer of the CNN network via training the feature maps in the SVM classifier. Next, taking the simulation experiments of MNIST handwritten digital dataset and UC Merced Land Use remote sensing dataset as examples, we tested and verified the proposed method in this experiment. Finally, the empirical study of volcanic ash cloud (VAC) classification from moderate resolution imaging spectroradiometer (MODIS) RSI was carried out and evaluated. The experimental results show that compared with the traditional methods, the proposed method has lower loss value and better generalization in modeling training; the total classification accuracy of VAC and Kappa coefficient reached 93.5 and 0.8502, respectively, and achieved preferable VAC identification and visual effects. It will enhance the classification accuracy to the massive remote sensing data.

The fully-supervised deep convolutional neural network (CNN) cannot detect the discriminant local information which is responsible for spatial transformations in high-resolution remote sensing images. To address the various types and missing labels of natural disasters, a new deep multi-instance convolutional neural network (DMCNN) model for disaster classification in high-resolution remote sensing image is presented in this article. Specifically, based on sample enhancement and atrous spatial pyramid pooling (ASPP), we first extract and integrate the features via the CNN structure to obtain the instance feature of bags in the image. Besides, introducing a prototype learning layer with distance measure, the instance features extracted from pretrained CNN are mapped into a series of prototype instance features with bag-level. And then all instance features from prototype and bag take part in disaster detection and image classification. Finally, we conduct extensive experiments on xBD dataset and discussions from qualitative and quantitative aspects. Experimental results show that the proposed DMCNN model achieves better classification accuracy of natural disaster from high-resolution remote sensing image compared to traditional CNNs, and effectively improves the disaster classification performance with weakly-supervised from high-resolution remote sensing image.

Target detection of small samples with a complex background is always difficult in the classification of remote sensing images. We propose a new small sample target detection method combining local features and a convolutional neural network (LF-CNN) with the aim of detecting small numbers of unevenly distributed ground object targets in remote sensing images. The k-nearest neighbor method is used to construct the local neighborhood of each point and the local neighborhoods of the features are extracted one by one from the convolution layer. All the local features are aggregated by maximum pooling to obtain global feature representation. The classification probability of each category is then calculated and classified using the scaled expected linear units function and the full connection layer. The experimental results show that the proposed LF-CNN method has a high accuracy of target detection and classification for hyperspectral imager remote sensing data under the condition of small samples. Despite drawbacks in both time and complexity, the proposed LF-CNN method can more effectively integrate the local features of ground object samples and improve the accuracy of target identification and detection in small samples of remote sensing images than traditional target detection methods.

Monitoring of volcanic ash cloud is conducive to the disaster prevention and mitigation and public safety. To tackle of large amount and various types of data and continuous changes of volcanic ash cloud monitoring, in this paper, a new long short term memory (LSTM) and cellular automaton (CA) (i.e., LSTM-CA) collaborative computing method for volcanic ash cloud diffusion is proposed via neural networks. Based on diffusion characteristics of volcanic ash cloud, a CA model of volcanic ash cloud in the three-dimensional spaces was first constructed. And then the constantly changing sequential characteristics of volcanic ash cloud was learned by LSTM neural network and further treated as the evolution rule of the CA diffusion model of volcanic ash cloud in three-dimensional space. Next, simulation experiments and analysis were conducted in terms of wind direction, wind speed, step size and the number of cell. Finally, the proposed LSTM-CA collaborative computing method was tested and verified in the actual Etna ash cloud diffusion case. The experimental results show that: (1) in the two-dimensional space, the proposed LSTM-CA method can obtain a good initial simulation effect of volcanic ash cloud diffusion, and the total accuracy of volcanic ash cloud identification reached 96.1%; (2) in the three-dimensional space, the proposed LSTM-CA method can exact simulate the horizontal and vertical diffusion trends of volcanic ash cloud; (3) the proposed LSTM-CA method can significantly reduce the modeling complexity of volcanic ash cloud and improve the calculation efficiency of spatiotemporal data. It seems to provide a new idea to identify and simulate the volcanic ash cloud in complex environments.

The severe haze pollution episode from December 1 to 10, 2013 occurred in Shanghai area, and seriously threatened human health and transportation safety. As regional air pollution, it is practical significance to the rapid monitoring and comprehensive analysis for haze treatment. Based on the moderate resolution imaging spectroradiometer (MODIS) and cloud-aerosol LiDAR with orthogonal polarization (CALIOP) data, via combing with meteorological features during the haze pollution period, the current study analyzed and explored the spatial distribution, aerosol vertical structure, source and migration path of haze pollution. The results showed that: (1) There are obvious aerosol layers over Shanghai area, aerosol optical depth (AOD) fluctuates frequently and has distinctive features such as complexity, regionality and seasonality. (2) The aerosol layer is mainly at the heights ranging from 0 to 2 km and composed mostly of spherical aerosol but there is also non-spherical aerosol. (3) The total backscattering intensity of aerosol is centralized at the heights ranging from 0 to 1 km, especially at the heights ranging from 0 to 0.5 km. (4) The haze pollution episode was affected by both specific meteorological factors and migration of the non-local particles in the surrounding areas and northern China.

The operation and maintenance management of telecom platforms puts more emphasis on operation and alarm monitoring and less on business flow, and it is difficult to have a seamless connection between operation and maintenance and sales. To improve the cooperative efficiency and the informatization level of telecom platform operation and maintenance, a new operation and maintenance system integrating Apriori association rules for a telecom platform is proposed. In this paper, the Apriori algorithm is firstly used to analyse the fault correlation of the operation and maintenance system of the telecommunications platform to get the alarm message. Then, the fault risk of the operation and maintenance platform is evaluated intelligently by the system‐business alarm causality model. Finally, based on the fixed end and the mobile end, the dynamic presentation of the system running state and alarm monitoring, device query and positioning, and online and offline inspection are implemented and designed, respectively. Experimental results show that the proposed operation and maintenance systems integrating Apriori association rules for the telecom platform innovate the cooperative mode between the operations and sales team, and the average time reduces to 3.9 minutes from 4.7 minutes and the risk forecasting accuracy increases from 8% to 26%. The assumed method can significantly improve the operational efficiency and intelligent level of the telecom platform’s system and business.