Clouds in optical remote sensing images cause spectral information change or loss, that affects image analysis and application. Therefore, cloud detection is of great significance. However, there are some shortcomings in current methods, such as the insufficient extendibility due to using the information of multiple bands, the intense extendibility due to relying on some manually determined thresholds, and the limited accuracy, especially for thin clouds or complex scenes caused by low-level manual features. Combining the above shortcomings and the requirements for efficiency in practical applications, we propose a light-weight deep learning cloud detection network based on DeeplabV3+ architecture and channel attention module (CD-AttDLV3+), only using the most common red–green–blue and near-infrared bands. In the CD-AttDLV3+ architecture, an optimized backbone network-MobileNetV2 is used to reduce the number of parameters and calculations. Atrous spatial pyramid pooling effectively reduces the information loss caused by multiple down-samplings while extracting multi-scale features. CD-AttDLV3+ concatenates more low-level features than DeeplabV3+ to improve the cloud boundary quality. The channel attention module is introduced to strengthen the learning of important channels and improve the training efficiency. Moreover, the loss function is improved to alleviate the imbalance of samples. For the Landsat-8 Biome set, CD-AttDLV3+ achieves the highest accuracy in comparison with other methods, including Fmask, SVM, and SegNet, especially for distinguishing clouds from bright surfaces and detecting light-transmitting thin clouds. It can also perform well on other Landsat-8 and Sentinel-2 images. Experimental results indicate that CD-AttDLV3+ is robust, with a high accuracy and extendibility.

In order to preserve the spatial and spectral information of the original panchromatic and multi-spectral images, this study designs a loss function suitable for pan-sharpening and a four-layer convolution neural network that could adequately extract spectral and spatial features from original source images. The major advantage of this study is that the designed loss function does not need the reference fused image, and then the proposed pan-sharpening method does not need to make the simulation data for training. This is the big difference from most existing pan-sharpening methods. Moreover, the loss function takes into account the characteristics of remote sensing images, including the spatial and spectral evaluation indicators. We also add the feature enhancement layer in convolution neural network, thus, the proposed four-layer network contains feature extraction, feature enhancement, linear mapping and reconstruction. In order to evaluate the effectiveness and universality of the proposed fusion model, we selected thousands of remote sensing images which include different sensors, different times and different land-cover types to make the training dataset. By evaluating the performance on the WorldView-2, Pleiades and Gaofen-1 experimental data, the results show that the proposed method achieves optimal performance in terms of both the subjective visual effect and the object assessment.The code is available at https://github.com/Zhangxi-Xiong/pan-sharpening

Road extraction from the high resolution remote sensing image is significant for the land planning, vehicle navigation, etc. The existing road extraction methods normally need many pre-processing and subsequent optimization steps. Therefore, an automatic road centerline extraction method based on the self-supervised learning framework for high resolution remote sensing image is proposed. This proposed method does not need to manually select training samples and other optimization steps. Firstly, the positive sample selection method combining the spectral and shape features is proposed to extract the road sample. Then, the one-class classifier framework is introduced and the random forest positive unlabeled learning classifier is constructed to get the posterior probability of the pixel belonging to road. The shape feature and the posterior probability are combined to form the final road network in the object-oriented way. Finally, the road centerline is obtained through the tensor voting algorithm. In order to verify the effectiveness of the proposed algorithm, high resolution remote sensing images and benchmark datasets are used to do experiments. The indexes of the completeness, the correctness and the detection quality are used for the quantitative accuracy evaluation. Compared with the supervised, the unsupervised and the one-class classification road extraction algorithms, this proposed algorithm achieves high accuracy and efficiency.For the deep learning method comparison, the deep learning method performs well in most cases especially in the complex urban area. However, the deep learning method needs a large number of samples and a long training time, and our SSLF does not need training samples.