Search

Simple Summary: The use of computer vision technology has improved the effectiveness of individual sheep identification, but existing methods face challenges such as large parameter sizes, slow recognition speeds, and difficult deployment. To address this issue, we have made improvements and optimizations based on the Retinaface face recognition model to create a balanced speed-accuracy sheep face recognition model. The optimized model has fewer parameters, simpler computations, faster inference speeds, and higher recognition accuracy, making it well-suited for deployment on resource-constrained edge devices. This research is expected to promote the application of deep learning-based sheep face recognition methods in production. The recognition of sheep faces based on computer vision has improved the efficiency and effectiveness of individual sheep identification, providing technical support for the development of smart farming. However, current recognition models have problems such as large parameter sizes, slow recognition speed, and difficult deployment. Therefore, this paper proposes an efficient and fast basic module called Eblock and uses it to build a lightweight sheep face recognition model called SheepFaceNet, which achieves the best balance between speed and accuracy. SheepFaceNet includes two modules: SheepFaceNetDet for detection and SheepFaceNetRec for recognition. SheepFaceNetDet uses Eblock to construct the backbone network to enhance feature extraction capability and efficiency, designs a bidirectional FPN layer (BiFPN) to enhance geometric location ability, and optimizes the network structure, which affects inference speed, to achieve fast and accurate sheep face detection. SheepFaceNetRec uses Eblock to construct the feature extraction network, uses ECA channel attention to improve the effectiveness of feature extraction, and uses multi-scale feature fusion to achieve fast and accurate sheep face recognition. On our self-built sheep face dataset, SheepFaceNet recognized 387 sheep face images per second with an accuracy rate of 97.75%, achieving an advanced balance between speed and accuracy. This research is expected to further promote the application of deep-learning-based sheep face recognition methods in production. [ABSTRACT FROM AUTHOR]

In rockfall hazard management, the investigation and detection of potential rockfall source areas on rock cliffs by remote-sensing-based susceptibility analysis are of primary importance. However, when the rockfall analysis results are used as feedback to the fieldwork, the irregular slope surface morphology makes it difficult to objectively locate the risk zones of hazard maps on the real slopes, and the problem of straightforward on-site visualization of rockfall susceptibility remains a research gap. This paper presents some of the pioneering studies on the augmented reality (AR) mapping of geospatial information from cyberspace within 2D screens to the physical world for on-site visualization, which directly recognizes the rock mass and superimposes corresponding rock discontinuities and rockfall susceptibility onto the real slopes. A novel method of edge-based tracking of the rock mass target for mobile AR is proposed, where the model edges extracted from unmanned aerial vehicle (UAV) structure-from-motion (SfM) 3D reconstructions are aligned with the corresponding actual rock mass to estimate the camera pose accurately. Specifically, the visually prominent edges of dominant structural planes were first explored and discovered to be a robust visual feature of rock mass for AR tracking. The novel approaches of visual-geometric synthetic image (VGSI) and prominent structural plane (Pro-SP) were developed to extract structural planes with identified prominent edges as 3D template models which could provide a pose estimation reference. An experiment verified that the proposed Pro-SP template model could effectively improve the edge tracking performance and quality, and this approach was relatively robust to the changes of sunlight conditions. A case study was carried out on a typical roadcut cliff in the Mentougou District of Beijing, China. The results validate the scalability of the proposed mobile AR strategy, which is applicable and suitable for cliff-scale fieldwork. The results also demonstrate the feasibility, efficiency, and significance of the geoinformation AR mapping methodology for on-site zoning and locating of potential rockfalls, and providing relevant guidance for subsequent detailed site investigation. [ABSTRACT FROM AUTHOR]