Jianping Zhao

Despite significant progress has been made in image deraining, most methods only independently handle a single type of rain degradation (e.g., rain streak, raindrop or nighttime rain), which limits the model's capacity to adapt to real-world dynamic degradations, especially on all-time autonomous driving. To advance this field, we introduce a new task: multiple-in-one image deraining, which aims to simultaneously address multiple types of rain degradation using a universal mix-trained model. To benchmark this task, we first construct a high-quality dataset based on driving scenarios, called MIO-Rain, which contains four patterns: daytime rain streak, daytime raindrop, nighttime rain streak and nighttime raindrop. Furthermore, we develop an effective scale-aware trident Transformer network (STTformer) for multiple-in-one image deraining. The proposed framework integrates coarse-to-fine and multi-patch parallel multi-branch architectures, with each branch employing multi-scale learning for its robustness to diverse rain appearances. Extensive experiments demonstrate the effectiveness of our model, and show that it achieves favorable performance against state-of-the-art ones.

Rain streaks in outdoor images can severely affect the performance of high-level vision tasks, such as surveillance and autonomous driving, by obscuring important image details. The objective of this research is to develop a more effective deraining method that preserves the structural details of images while removing rain streaks. To address the limitations of existing methods, we introduce the Recursive Multi-Scale Network with Enhanced Attention for Structure-Preserving Deraining (RMASP-Net). This network integrates multi-scale feature processing, recursive architecture, attention mechanisms, and Rectified Local Contrast Normalization (RLCN). It utilizes wavelet transform’s scale-space capabilities to detect key features across different scales, improving the alignment of structural features and isolating rain pixels effectively. Additionally, RMASP-Net features a Mask-Guided Attention Module (Mask-GAM) for dynamic feature weighting and an Interactive Fusion Module (IFM) to enhance image clarity and structure. Extensive experiments on both synthetic and real-world datasets demonstrate that RMASP-Net significantly outperforms existing methods. Specifically, RMASP-Net shows noticeable improvements in both PSNR and SSIM across datasets like Rain100H, Rain100L, and Test1200, providing clearer and more structurally accurate derained images. This research sets a new benchmark for deraining performance under real-world conditions.

Monocular panoramic depth estimation has various applications in robotics and autonomous driving due to its ability to perceive the entire field of view. However, panoramic depth estimation faces two significant challenges: global context capturing and distortion awareness. In this paper, we propose a new framework for panoramic depth estimation that can simultaneously address panoramic distortion and extract global context information, thereby improving the performance of panoramic depth estimation. Specifically, we introduce an attention mechanism into the multi-scale dilated convolution and adaptively adjust the receptive field size between different spatial positions, designing the adaptive attention dilated convolution module, which effectively perceives distortion. At the same time, we design the global scene understanding module to integrate global context information into the feature maps generated using the feature extractor. Finally, we trained and evaluated our model on three benchmark datasets which contains the virtual and real-world RGB-D panorama datasets. The experimental results show that the proposed method achieves competitive performance, comparable to existing techniques in both quantitative and qualitative evaluations. Furthermore, our method has fewer parameters and more flexibility, making it a scalable solution in mobile AR.

Element extraction of Network Security Situation is the foundation for Network Security Situation awareness. Traditional methods for network security element extraction struggle with efficiently processing large-scale data and establishing effective cross-level element associations, leading to a higher false alarm rate in situation awareness. This paper proposes a cross-level network security element fusion extraction method based on deep learning. Firstly, through a data preprocessing method based on element correlation, we reduce data dimensionality, addressing the challenge of large-scale nonlinear data processing. Secondly, we employ principal component analysis to construct a method for cross-level network security element fusion extraction. Moreover, by integrating convolutional neural networks and long short-term memory networks, we establish a cross-level network security element fusion classification model, achieving precise perception of the Network Security Situation. Experimental results indicate that the proposed cross-level network security element fusion extraction method effectively reduces dataset dimensions, extracts 32-dimensional crucial network security fusion elements, and achieves a detection accuracy rate of up to 99.99% for network attack behaviors, significantly enhancing the capability of network security situation awareness.

To solve the problems of low convergence accuracy, slow speed, and common falls into local optima of the Chicken Swarm Optimization Algorithm (CSO), a performance enhancement strategy of the CSO algorithm (PECSO) is proposed with the aim of overcoming its deficiencies. Firstly, the hierarchy is established by the free grouping mechanism, which enhances the diversity of individuals in the hierarchy and expands the exploration range of the search space. Secondly, the number of niches is divided, with the hen as the center. By introducing synchronous updating and spiral learning strategies among the individuals in the niche, the balance between exploration and exploitation can be maintained more effectively. Finally, the performance of the PECSO algorithm is verified by the CEC2017 benchmark function. Experiments show that, compared with other algorithms, the proposed algorithm has the advantages of fast convergence, high precision and strong stability. Meanwhile, in order to investigate the potential of the PECSO algorithm in dealing with practical problems, three engineering optimization cases and the inverse kinematic solution of the robot are considered. The simulation results indicate that the PECSO algorithm can obtain a good solution to engineering optimization problems and has a better competitive effect on solving the inverse kinematics of robots.

As a further investigation of the image captioning task, some works extended the vision-text dataset for specific subtasks, such as the stylized caption generating. The corpus in such dataset is usually composed of obvious sentiment-bearing words. While, in some special cases, the captions are classified depending on image category. This will result in a latent problem: the generated sentences are in close semantic meaning but belong to different or even opposite categories. It is a worthy issue to explore an effective way to utilize the image category label to boost the caption difference. Therefore, we proposed an image captioning network with the label control mechanism (LCNET) in this paper. First, to further improve the caption difference, LCNET employs a semantic enhancement module to provide the decoder with global semantic vectors. Then, through the proposed label control LSTM, LCNET can dynamically modulate the caption generation depending on the image category labels. Finally, the decoder integrates the spatial image features with global semantic vectors to output the caption. Using all the standard evaluation metrics shows that our model outperforms the compared models. Caption analysis demonstrates our approach can improve the performance of semantic representation. Compared with other label control mechanisms, our model is capable of boosting the caption difference according to the labels and keeping a better consistent with image content as well.