Animal behavior monitoring allows the gathering of animal health information and living habits and is an important technical means in precision animal farming. To quickly and accurately identify the behavior of broilers at different days, we adopted different deep learning behavior recognition models. Firstly, the top-view images of broilers at 2, 9, 16 and 23 days were obtained. In each stage, 300 images of each of the four broilers behaviors (i.e., feeding, drinking, standing, and resting) were segmented, totaling 4800 images. After image augmentation processing, 10,200 images were generated for each day including 8000 training sets, 2000 validation sets, and 200 testing sets. Finally, the performance of different convolutional neural network models (CNN) in broiler behavior recognition at different days was analyzed. The results show that the overall performance of the DenseNet-264 network was the best, with the accuracy rates of 88.5%, 97%, 94.5%, and 90% when birds were 2, 9, 16 and 23 days old, respectively. In addition, the efficient channel attention was introduced into the DenseNet-264 network (ECA-DenseNet-264), and the results (accuracy rates: 85%, 95%, 92%, 89.5%) confirmed that the DenseNet-264 network was still the best overall. The research results demonstrate that it is feasible to apply deep learning technology to monitor the behavior of broilers at different days.

Animal tracking and feeding monitoring is crucial for automatic individual cow welfare measurement and naturally becomes a prerequisite for autonomous livestock farming systems. The deformable body posture and irregular movement of cows under complex farming environments make tracking of individual animals in a herd very challenging. To tackle the above challenge, a deep learning network-based approach, namely, YOLOv5s-CA+DeepSORT-ViT , is proposed in this article. In our proposed approach, coordinate attention (CA)-integrated YOLOv5 was developed to capture spatial location information to improve the face detection performance for overlapping regions. Then the vision transformer (ViT) was embedded in the reidentification (reID) network Deep Simple Online and Real-time Tracking (DeepSORT) to enhance feature matching and tracking accuracy. The comparative results of the multicow complex dataset constructed from a commercial farm show that the ID F1 score (IDF1) and multitarget tracking accuracy (MOTA) of the proposed YOLOv5s-CA+DeepSORT-ViT are 88.5% and 84.4%, respectively. Meanwhile, the ID switching (ID Sw.) times and the processing time are reduced by 50% and 20% compared to the YOLOv5s+DeepSORT model. Experimental results also showed that the overall cow tracking performance of our proposed approach outperformed the other baselines (e.g. SORT, ByteTrack, BoT-SORT, and DeepSORT).

With the advancement of agricultural intelligence, dairy-cow farming has become a significant industry, and the application of computer vision technology in the automated monitoring of dairy cows has also attracted much attention. However, most of the images in the conventional detection dataset are high-quality images under normal lighting, which makes object detection very challenging in low-light environments at night. Therefore, this study proposed a night-time detection framework for cows based on an improved lightweight Zero-DCE (Zero-Reference Deep Curve Estimation) image enhancement network for low-light images. Firstly, the original feature extraction network of Zero-DCE was redesigned with an upsampling structure to reduce the influence of noise. Secondly, a self-attention gating mechanism was introduced in the skip connections of the Zero-DCE to enhance the network’s attention to the cow area. Then, an improved kernel selection module was introduced in the feature fusion stage to adaptively adjust the size of the receptive field. Finally, a depthwise separable convolution was used to replace the standard convolution of Zero-DCE, and an Attentive Convolutional Transformer (ACT) module was used to replace the iterative approach in Zero-DCE, which further reduced the computational complexity of the network and speeded up the inference. Four different object-detection models, YOLOv5, CenterNet, EfficientDet, and YOLOv7-tiny, were selected to evaluate the performance of the improved network and were tested on the night-time dataset before and after enhancement. Experimental results demonstrate that the detection performance of all models is significantly improved when processing night-time image samples through the enhanced Zero-DCE model. In summary, the improved lightweight Zero-DCE low-light enhancement network proposed in this study shows excellent performance, which can ensure that various object-detection models can quickly and accurately identify targets in low-light environments at night and are suitable for real-time monitoring in actual production environments.