Rotation-Invariant Attention Network for Hyperspectral Image Classification

UCFNNet: Ulcerative colitis evaluation based on fine-grained lesion learner and noise suppression gating

BACKGROUND AND OBJECTIVE

Ulcerative colitis (UC) is a chronic disease characterized by recurrent symptoms and significant morbidity. The exact cause of the disease remains unknown. The selection of current treatment options for ulcerative colitis depends on the severity and location of the disease in each patient. Therefore, developing a fully automated endoscopic images for evaluating UC is crucial for guiding treatment plans and facilitating early prevention efforts.

METHODS

We propose a network called ulcerative colitis evaluation based on fine-grained lesion learner and noise suppression gating (UCFNNet). UCFNNet contains three novel modules. Firstly, a fine-grained lesion feature learner (FG-LF Learner) is proposed by integrating local features and a Softmax category prediction (SCP) module to improve the feature accuracy in small lesion areas. Subsequently, a graph convolutional feature combiner (GCFC) is developed to connect features across adjacent convolutional layers and to incorporate short connections between input and output, thereby mitigating feature loss during transmission. Thereafter, a noise suppression gating (NS gating) technique is designed by implementing a grid attention mechanism and a feature gating (FG) module to prioritize significant lesion features and suppress irrelevant and noisy regions in the input feature map.

RESULTS

We evaluate the performance of the proposed network on both privately-collected and publicly-available datasets. The evaluation of UC achieves excellent results on privately-collected dataset, with an accuracy (ACC) of 89.57 %, Matthews correlation coefficient (MCC) of 85.52 %, precision of 89.26 %, recall of 89.48 %, and F1-score of 89.78 %. The results are also impressive on publicly-available dataset, with ACC of 85.47 %, MCC of 80.42 %, precision of 85.62 %, recall of 84.00 %, and F1-score of 84.53 %, surpassing the performance of state-of-the-art techniques.

CONCLUSION

Our proposed model introduces three innovative algorithm modules, which outperform the current state-of-the-art methods and achieve high ACC and F1-score. This indicates that our method has superior performance compared to traditional machine learning and existing deep methods, which means that our method has good application prospects. Meanwhile, it has been verified that the proposed model demonstrates good interpretability. The source code is available at github.com/YinLeRenNB/UCFNNet.

A Deep Stochastic Adaptive Fourier Decomposition Network for Hyperspectral Image Classification

Deep learning-based hyperspectral image (HSI) classification methods have recently shown excellent performance, however, there are two shortcomings that need to be addressed. One is that deep network training requires a large number of labeled images, and the other is that deep network needs to learn a large number of parameters. They are also general problems of deep networks, especially in applications that require professional techniques to acquire and label images, such as HSI and medical images. In this paper, we propose a deep network architecture (SAFDNet) based on the stochastic adaptive Fourier decomposition (SAFD) theory. SAFD has powerful unsupervised feature extraction capabilities, so the entire deep network only requires a small number of annotated images to train the classifier. In addition, we use fewer convolution kernels in the entire deep network, which greatly reduces the number of deep network parameters. SAFD is a newly developed signal processing tool with solid mathematical foundation, which is used to construct the unsupervised deep feature extraction mechanism of SAFDNet. Experimental results on three popular HSI classification datasets show that our proposed SAFDNet outperforms other compared state-of-the-art deep learning methods in HSI classification.

CEGAT: A CNN and enhanced-GAT based on key sample selection strategy for hyperspectral image classification

In recent years, the application of convolutional neural networks (CNNs) and graph convolutional networks (GCNs) in hyperspectral image classification (HSIC) has achieved remarkable results. However, the limited label samples are still a major challenge when using CNN and GCN to classify hyperspectral images. In order to alleviate this problem, a double branch fusion network of CNN and enhanced graph attention network (CEGAT) based on key sample selection strategy is proposed. First, a linear discrimination of spectral inter-class slices (LD_SICS) module is designed to eliminate spectral redundancy of HSIs. Then, a spatial spectral correlation attention (SSCA) module is proposed, which can extract and assign attention weight to the spatial and spectral correlation features. On the graph attention (GAT) branch, the HSI is segmented into some super pixels as input to reduce the amount of network parameters. In addition, an enhanced graph attention (EGAT) module is constructed to enhance the relationship between nodes. Finally, a key sample selection (KSS) strategy is proposed to enable the network to achieve better classification performance with few labeled samples. Compared with other state-of-the-art methods, CEGAT has better classification performance under limited label samples.

Improved YOLOv5-based for small traffic sign detection under complex weather

Traffic sign detection is a challenging task for unmanned driving systems. In the traffic sign detection process, the object size and weather conditions vary widely, which will have a certain impact on the detection accuracy. In order to solve the problem of balanced detecting precision of traffic sign recognition model in different weather conditions, and it is difficult to detect occluded objects and small objects, this paper proposes a small object detection algorithm based on improved YOLOv5s in complex weather. First, we add the coordinate attention(CA) mechanism in the backbone, a light-weight yet effective module, embedding the location information of traffic signs into the channel attention to improve the feature extraction ability of the network. Second, we exploit effectively fine-grained features about small traffic signs from the shallower layers by adding one prediction head to YOLOv5s. Finally, we use Alpha-IoU to improve the original positioning loss CIoU, improving the accuracy of bbox regression. Applying this model to the recently proposed CCTSDB 2021 dataset, for small objects, the precision is 88.1%, and the recall rate is 79.8%, compared with the original YOLOv5s model, it is improved by 12.5% and 23.9% respectively, and small traffic signs can be effectively detected under different weather conditions, with low miss rate and high detection accuracy. The source code will be made publicly available at https://github.com/yang-0706/ImprovedYOLOv5s .

A Tracking Imaging Control Method for Dual-FSM 3D GISC LiDAR

In this paper, a tracking and pointing control system with dual-FSM (fast steering mirror) composite axis is proposed. It is applied to the target-tracking accuracy control in a 3D GISC LiDAR (three-dimensional ghost imaging LiDAR via sparsity constraint) system. The tracking and pointing imaging control system of the dual-FSM 3D GISC LiDAR proposed in this paper is a staring imaging method with multiple measurements, which mainly solves the problem of high-resolution remote-sensing imaging of high-speed moving targets when the technology is transformed into practical applications. In the research of this control system, firstly, we propose a method that combines motion decoupling and sensor decoupling to solve the mechanical coupling problem caused by the noncoaxial sensor installation of the FSM. Secondly, we suppress the inherent mechanical resonance of the FSM in the control system. Thirdly, we propose the optical path design of a dual-FSM 3D GISC LiDAR tracking imaging system to solve the problem of receiving aperture constraint. Finally, after sufficient experimental verification, our method is shown to successfully reduce the coupling from 7% to 0.6%, and the precision tracking bandwidth reaches 300 Hz. Moreover, when the distance between the GISC system and the target is 2.74 km and the target flight speed is 7 m/s, the tracking accuracy of the system is improved from 15.7 μrad (σ) to 2.2 μrad (σ), and at the same time, the system recognizes the target contour clearly. Our research is valuable to put the GISC technology into practical applications.