The Chengdu ma goat is an excellent local breed in China. As one of the breeds listed in the National List of Livestock and Poultry Genetic Resources Protection, the protection of its germplasm resources is particularly important. However, the existing breeding and protection methods for them are relatively simple, due to the weak technical force and lack of intelligent means to assist. Most livestock farmers still conduct small-scale breeding in primitive ways, which is not conducive to the breeding and protection of Chengdu ma goats. In this paper, an automatic individual recognition method for Chengdu ma goats is proposed, which saves labor costs and does not depend on large-scale mechanized facilities. The main contributions of our work are as follows: (1) a new Chengdu ma goat dataset is built, which forms the basis for object detection and classification tasks; (2) an improved detection algorithm for Chengdu ma goats based on TPH-YOLOv5 is proposed, which is able to accurately localize goats in high-density scenes with severe scale variance of targets; (3) a classifier incorporating a self-supervised learning module is implemented to improve the classification performance without increasing the labeled data and inference computation overhead. Experiments show that our method is able to accurately recognize Chengdu ma goats in the actual indoor barn breeding environment, which lays the foundation for precision feeding based on sex and age.

Tree species classification is an important and challenging task in image recognition and the management of forest resources. Moreover, the task of tree species classification based on remote sensing images can significantly improve the efficiency of the tree species survey and save costs. In recent years, many large models have achieved high accuracy in the task of tree species classification in an airborne remote-sensing manner, but due to their fixed geometric structure, traditional convolutional neural networks are inherently limited to the local receptive field and can only provide segmental context information. The limitation of insufficient context information greatly affects the segmentation accuracy. In this paper, a dual-attention residual network (AMDNet) and a re-parameterized model approach are proposed to capture the global context information, fuse the weight, reduce the model volume, and maintain the computational efficiency. Firstly, we propose MobileNetV2 as the backbone network for feature extraction to further improve the feature identification by modeling semantic dependencies in the spatial dimension and channel dimension and adding the output of the two attention modules. Then, the attention perception features are generated by stacking the attention modules, and the in-depth residual attention network is trained using attention residual learning, through which more accurate segmentation results can be obtained. Secondly, we adopt the approach of structure re-parameterization, use a multi-branch topology for training, carry out weighted averaging on multiple trained models, and fuse multiple branch modules into a completely equivalent module in inference. The proposed approach results in a reduction in the number of parameters and an accelerated inference speed while also achieving improved classification accuracy. In addition, the model training strategy is optimized based on Transformer to enhance the accuracy of segmentation. The model was used to conduct classification experiments on aerial orthophotos of Hongya Forest Farm in Sichuan, China, and the MIOU of tree species recognition using the test equipment reached 93.8%. Compared with current models such as UNet, our model exhibits a better performance in terms of both speed and accuracy, in addition to its enhanced deployment capacity, and its speed advantage is more conducive to real-time segmentation, thereby representing a novel approach for the classification of tree species in remote sensing imagery with significant potential for practical applications.

The slow loris (Genus Nycticebus) is a group of small, nocturnal and venomous primates with a distinctive locomotion mode. The detection of slow loris plays an important role in the subsequent individual identification and behavioral recognition and thus contributes to formulating targeted conservation strategies, particularly in reintroduction and post-release monitoring. However, fewer studies have been conducted on efficient and accurate detection methods of this endangered taxa. The traditional methods to detect the slow loris involve long-term observation or watching surveillance video repeatedly, which would involve manpower and be time consuming. Because humans cannot maintain a high degree of attention for a long time, they are also prone to making missed detections or false detections. Due to these observational challenges, using computer vision to detect slow loris presence and activity is desirable. This article establishes a novel target detection dataset based on monitoring videos of captive Bengal slow loris (N. bengalensis) from the wildlife rescue centers in Xishuangbanna and Pu’er, Yunnan, China. The dataset is used to test two improvement schemes based on the YOLOv5 network: (1) YOLOv5-CBAM + TC, the attention mechanism and deconvolution are introduced; (2) YOLOv5-SD, the small object detection layer is added. The results demonstrate that the YOLOv5-CBAM + TC effectively improves the detection effect. At the cost of increasing the model size by 0.6 MB, the precision rate, the recall rate and the mean average precision (mAP) are increased by 2.9%, 3.7% and 3.5%, respectively. The YOLOv5-CBAM + TC model can be used as an effective method to detect individual slow loris in a captive environment, which helps to realize slow loris face and posture recognition based on computer vision.