Issues such as insufficient feature representation and background confusion make detection tasks for small object in remote sensing arduous. Particularly when the algorithm will be deployed on board for real-time processing, which requires extensive optimization of accuracy and speed under limited computing resources. To tackle these problems, an efficient detector called FFCA-YOLO(Feature enhancement, Fusion and Context Aware YOLO) is proposed in this paper. FFCA-YOLO includes three innovative lightweight and plug-and-play modules: feature enhancement module(FEM), feature fusion module(FFM) and spatial context aware module(SCAM). These three modules improve the network capabilities of local area awareness, multi-scale feature fusion and global association cross channels and space, respectively, while trying to avoid increasing complexity as possible. Thus the weak feature representations of small objects are enhanced and the confusable backgrounds are suppressed. Two public remote sensing datasets(VEDAI and AI-TOD) for small object detection and one self-built dataset(USOD) are used to validate the effectiveness of FFCA-YOLO. The accuracy of FFCA-YOLO reaches 0.748, 0.617 and 0.909(in terms of mAP ₅₀ ) that exceeds several benchmark models and state-of-the-art methods. Meanwhile, the robustness of FFCA-YOLO is also validated under different simulated degradation conditions. Moreover, to further reduce computational resource consumption while ensuring efficiency, a lite version of FFCA-YOLO(L-FFCA-YOLO) is optimized by reconstructing the backbone and neck of FFCA-YOLO based on partial convolution. L-FFCA-YOLO has faster speed, smaller parameter scale, lower computing power requirement but little accuracy loss compared with FFCA-YOLO. The source code will be available at https://github.com/yemu1138178251/FFCA-YOLO.

Driver fatigue is the culprit of most traffic accidents. Visual technology can intuitively judge whether the driver is in the state of fatigue. A driver fatigue detection system based on the residual channel attention network (RCAN) and head pose estimation is proposed. In the proposed system, Retinaface is employed for face location and outputs five face landmarks. Then the RCAN is proposed to classify the state of eyes and the mouth. The RCAN includes a channel attention module, which can adaptively extract key feature vectors from the feature map, which significantly improves the classification accuracy of the RCAN. In the self-built dataset, the classification accuracy of the eye state of the RCAN reaches 98.962% and that of the mouth state reaches 98.561%, exceeding other classical convolutional neural networks. The percentage of eyelid closure over the pupil over time (PERCLOS) and the mouth opening degree (POM) are used for fatigue detection based on the state of eyes and the mouth. In addition, this article proposes to use a Perspective-n-Point (PnP) method to estimate the head pose as an essential supplement for driving fatigue detection and proposes over-angle to evaluate whether the head pose is excessively deflected. On the whole, the proposed driver fatigue system integrates 3D head pose estimation and fatigue detection based on deep learning. This system is evaluated by the four datasets and shows success of the proposed method with their high performance.

In remote sensing images, vehicles often appear on a minuscule scale, lacking features and easily overwhelmed by intricate background information. This becomes even more challenging in low illumination or occluded environments, leading to missed detections and false alarms. A novel vehicle detection algorithm, known as Local and Global Aware YOLO (LGA-YOLO), is introduced to tackle these issues. LGA-YOLO incorporates two innovative and plug-and-play modules: the Multi-Scale Large Kernel Local Aware Module (MLKM) and the Directional Global Context Aware Module (DGAM). MLKM widens the receptive field and enhances local features, while DGAM gathers global context information, highlighting vehicle features against complex backgrounds. Based on these modules, a high-low feature fusion network is reconstructed, capturing multi-scale object features and effectively leveraging shallow features. Our self-constructed dataset (USOD), VEDAI, and DOTA are employed to validate LGA-YOLO's efficacy. In USOD, the results demonstrate the remarkable performance of LGA-YOLO, with precision, recall, AP0.5, and AP0.5:0.95 scores of 0.927, 0.889, 0.930, and 0.371, respectively. In VEDAI and DOTA, the mAP0.5 of LGA-YOLO reaches 0.803 and 0.781, respectively. These metrics not only surpass baseline models but also leading-edge algorithms in the field.

In industrial vision system, metal surface is anisotropic under light in all directions and it is inevitable to cause local overexposure due to the natural reflection of active strong light, especially on the cylindrical metal surface. In this paper, injector valve is taken as the representative of cylindrical metal workpieces. Since the variety and complexity of cylindrical metal workpieces defects, and its contrast with the background of workpieces fluctuates, making samples annotating time-consuming and be of high cost. In order to solve the above challenges, this paper proposes an end-to-end weakly supervised learning framework to classify and segment defects. Firstly, a deep integrated residual attention convolutional neural network (IRA-CNN) is designed. IRA-CNN is composed of multiple IRA-Block. Two residual maps are included in IRA-Block to improve its bilateral nonlinearity and the robustness. IRA-block adds integrated attention module (IAM) which includes channel attention submodule and spatial attention submodule. The channel attention submodule adaptively extracts information from the global average pool layer and the global maximum pool layer to obtain the channel attention feature map. IAM can be well integrated into the IRA-CNN makes the neural network suppress the interference of useless background area and highlight the defect area. Finally, the weakly supervised segmentation method relies on Grad-CAM++ to generate saliency map to improve segmentation accuracy. The experimental results show that the accuracy of defect classification reaches 97.7\% and the segmentation accuracy is significantly improved compared with the benchmark method in the injector valve dataset which include 6747 images.