Body condition scores (BCS) is an important parameter, which is in high correlation with the health status of a dairy cow, metabolic disorder and milk composition during the production period. To evaluate BCS, the traditional methods rely on veterinary experts or skilled staff to look at a cow and touch it. These methods have low efficiency especially on large-scale farms. Computer vision methods are widely used but there are some improvements to increase BCS accuracy. In this study, a low cost BCS evaluation method based on deep learning and machine vision is proposed. Firstly, the back-view images of the cows are captured by network cameras, resulting in 8972 images that constituted the sample data set. The camera is a common 2D camera, which is cheaper and easier to install compared with 3D cameras. Secondly, the key body parts such as tails, pins and rump in the images were labeled manually, the Sing Shot multi-box Detector (SSD) method was used to detect the tail and evaluate the BCS. Inspired by DenseNet and Inception-v4, a new SSD was introduced by changing the network connection method of the original SSD. Finally, the experiments show that the improved SSD method can achieve 98.46% classification accuracy and 89.63% location accuracy, and it has: (1) faster detection speed with 115 fps; (2) smaller model size with 23.1 MB compared to original SSD and YOLO-v3, these are significant advantages for reducing hardware costs.

Cow tail detection and tracking in videos provide valuable information for individual identification, calving process, behavior analysis, and body condition monitoring. Although deep learning-based detection methods have demonstrated good performance, many of them have high complexity and still require an improvement in video object detection by reducing the computation time and false positives. To fully exploit the interframe information and achieve a real-time online detection, the optimized cow tail detection and tracking method is proposed based on an improved single shot multibox detector (SSD) and Kalman filter. Here, our improved SSD-integrated DenseNet and Inception-v4 reduce detection information loss and network parameters. Then, an improved window function-based Kalman filter and Hungarian are adopted to remove error detections and enhance the cow tail tracking accuracy. Experiments on our acquired rear-view videos show that the proposed approach achieved fast tail detection with an accuracy of 96.97%, a speed of 96 fps, a smaller model size of 25 MB, and higher position accuracy with an average 6.45 pixels deviation. The proposed approach outperformed the region-based R-CNN models and other tracking methods (e.g., particle filter), which provides a new solution to automatic cow detection and tracking in smart livestock farming.

Posture recognition has been widely applied in fields such as physical training, environmental awareness, human-computer-interaction, surveillance system and elderly health care. The traditional methods consist of two main variations: machine vision methods and acceleration sensor methods. The former has the disadvantages of privacy invasion, high cost and complex implementation processes, while the latter has low recognition rate for still postures. A new body posture recognition scheme based on indoor positioning technology is presented in this paper. A single deployed indoor positioning system is constructed by installing wearable receiving tags at key points of the human body. The distance measurement method with ultra-wide band (UWB) radio is applied to position the key points of human body. Posture recognition is implemented by positioning. In the posture recognition algorithm, least square estimation (LSE) method and the improved extended Kalman filtering (iEKF) algorithm are respectively adopted to suppress the noise of the distances measurement and to improve the accuracy of positioning and recognition. The comparison of simulation results with the two methods shows that the improved extended Kalman filtering algorithm is more effective in error performance.

Obtaining animal regions and the relative position relationship of animals in the scene is conducive to further studying animal habits, which is of great significance for smart animal farming. However, the complex breeding environment still makes detection difficult. To address the problems of poor target segmentation effects and the weak generalization ability of existing semantic segmentation models in complex scenes, a semantic segmentation model based on an improved DeepLabV3+ network (Imp-DeepLabV3+) was proposed. Firstly, the backbone network of the DeepLabV3+ model was replaced by MobileNetV2 to enhance the feature extraction capability of the model. Then, the layer-by-layer feature fusion method was adopted in the Decoder stage to integrate high-level semantic feature information with low-level high-resolution feature information at multi-scale to achieve more precise up-sampling operation. Finally, the SENet module was further introduced into the network to enhance information interaction after feature fusion and improve the segmentation precision of the model under complex datasets. The experimental results demonstrate that the Imp-DeepLabV3+ model achieved a high pixel accuracy (PA) of 99.4%, a mean pixel accuracy (MPA) of 98.1%, and a mean intersection over union (MIoU) of 96.8%. Compared to the original DeepLabV3+ model, the segmentation performance of the improved model significantly improved. Moreover, the overall segmentation performance of the Imp-DeepLabV3+ model surpassed that of other commonly used semantic segmentation models, such as Fully Convolutional Networks (FCNs), Lite Reduced Atrous Spatial Pyramid Pooling (LR-ASPP), and U-Net. Therefore, this study can be applied to the field of scene segmentation and is conducive to further analyzing individual information and promoting the development of intelligent animal farming.