With the development of deep learning in remote sensing image change detection, the dependence of change detection models on labeled data has become an important problem. To make better use of the comparatively resource-saving unlabeled data, the change detection method based on semi-supervised learning is worth further study. This paper proposes a reliable contrastive learning method for semi-supervised remote sensing image change detection. First, according to the task characteristics of change detection, we design the contrastive loss based on the changed areas to enhance the model’s feature extraction ability for changed objects. Then, to improve the quality of pseudo labels in semi-supervised learning, we use the uncertainty of unlabeled data to select reliable pseudo labels for model training. Combining these methods, semi-supervised change detection models can make full use of unlabeled data. Extensive experiments on three widely used change detection datasets demonstrate the effectiveness of the proposed method. The results show that our semi-supervised approach has better performance than related methods. The code is available at https://github.com/VCISwang/RC-Change-Detection.

Short-term precipitation forecast in local areas based on radar reflectance images has become a hot spot issue in the meteorological field, which has an important impact on daily life. Recently, deep learning techniques have been applied to this field, and the effect is promoted remarkably compared with traditional methods. However, existing deep learning-based methods have not considered the problem that different areas and channels exert different influence on precipitation. In this paper, we propose to incorporate the multihead attention into a dual-channel neural network to highlight the key areas for precipitation forecast. Furthermore, to solve the problem of excessive loss of global information caused by the attention mechanism, the residual connection is introduced into the proposed model. Quantitative and qualitative results demonstrate that the proposed method achieves the state-of-the-art precipitation forecast accuracy on the radar echo dataset.

Current visual-based simultaneous localization and mapping(SLAM) system suffers from feature loss caused by fast motion and unstructured scene in complex environments. Addressing this problem, a fast semi-direct SLAM algorithm is proposed in this paper. The main idea of this method is to combine the feature point method with the direct method in order to improve the robustness of the system in the environment of scarce visual features and low texture. First, the feature enhancement module based on subgraph is developed to extract image feature points more stably. Second, an apparent shape weighted fusion method is proposed for camera pose estimation, which can still work robustly in the absence of feature points. Third, an incremental dynamic covariance scaling algorithm is studied for optimizing the error of camera pose estimation. Finally, based on the optimized camera pose, a face element model is designed to estimate and fuse the point cloud pose, and obtain an ideal three-dimensional point cloud map. The proposed algorithm has been tested extensively on the benchmark TUM dataset and the real environment. The results show that the algorithm has better performance than existing visual based SLAM algorithms.

Neural network quantization has shown to be an effective way for network compression and acceleration. However, existing binary or ternary quantization methods suffer from two major issues. First, low bit-width input/activation quantization easily results in severe prediction accuracy degradation. Second, network training and quantization are always treated as two non-related tasks, leading to accumulated parameter training error and quantization error. In this work, we introduce a novel scheme, named Residual Quantization, to train a neural network with both weights and inputs constrained to low bit-width, e.g., binary or ternary values. On one hand, by recursively performing residual quantization, the resulting binary/ternary network is guaranteed to approximate the full-precision network with much smaller errors. On the other hand, we mathematically re-formulate the network training scheme in an EM-like manner, which iteratively performs network quantization and parameter optimization. During expectation, the low bit-width network is encouraged to approximate the full-precision network. During maximization, the low bit-width network is further tuned to gain better representation capability. Extensive experiments well demonstrate that the proposed quantization scheme outperforms previous low bit-width methods and achieves much closer performance to the full-precision counterpart.