Dairy cow behavior carries important health information. Timely and accurate detection of behaviors such as drinking, feeding, lying, and standing is meaningful for monitoring individual cows and herd management. In this study, a model called Res-DenseYOLO is proposed for accurately detecting the individual behavior of dairy cows living in cowsheds. Specifically, a dense module was integrated into the backbone network of YOLOv5 to strengthen feature extraction for actual cowshed environments. A CoordAtt attention mechanism and SioU loss function were added to enhance feature learning and training convergence. Multi-scale detection heads were designed to improve small target detection. The model was trained and tested on 5516 images collected from monitoring videos of a dairy cowshed. The experimental results showed that the performance of Res-DenseYOLO proposed in this paper is better than that of Fast-RCNN, SSD, YOLOv4, YOLOv7, and other detection models in terms of precision, recall, and mAP metrics. Specifically, Res-DenseYOLO achieved 94.7% precision, 91.2% recall, and 96.3% mAP, outperforming the baseline YOLOv5 model by 0.7%, 4.2%, and 3.7%, respectively. This research developed a useful solution for real-time and accurate detection of dairy cow behaviors with video monitoring only, providing valuable behavioral data for animal welfare and production management.

In real-world dairy farming environments, object recognition models often suffer from missed or false detections due to complex backgrounds and cow occlusions. In response to these issues, this paper proposes FSCA-YOLO, a multi-object cow behavior recognition model based on an improved YOLOv11 framework. First, the FEM-SCAM module is introduced along with the CoordAtt mechanism to enable the model to better focus on effective behavioral features of cows while suppressing irrelevant background information. Second, a small object detection head is added to enhance the model’s ability to recognize cow behaviors occurring at the distant regions of the camera’s field of view. Finally, the original loss function is replaced with the SIoU loss function to improve recognition accuracy and accelerate model convergence. Experimental results show that compared with mainstream object detection models, the improved YOLOv11 in this section demonstrates superior performance in terms of precision, recall, and mean average precision (mAP), achieving 95.7% precision, 92.1% recall, and 94.5% mAP—an improvement of 1.6%, 1.8%, and 2.1%, respectively, over the baseline YOLOv11 model. FSCA-YOLO can accurately extract cow features in real farming environments, providing a reliable vision-based solution for cow behavior recognition. To support specific behavior recognition and in-region counting needs in multi-object cow behavior recognition and tracking systems, OpenCV is integrated with the recognition model, enabling users to meet the diverse behavior identification requirements in groups of cows and improving the model’s adaptability and practical utility.

Dry matter intake (DMI) in dairy cows is a critical indicator of nutrient intake from feed, serving as the cornerstone of precision feeding practices, playing a critical role in improving production efficiency and enhancing the quality of dairy products. To address the high costs of traditional measurement methods and the structural complexity and large parameter counts of neural network models, this study proposes a Stacking ensemble learning model to assess DMI, with model parameters optimized using the Tuna Swarm Optimization (TSO) algorithm to enhance assessment accuracy, taking cow body weight, lying duration, lying times, rumination duration, foraging duration, walking steps, and the concentrate-to-roughage feed ratio as input variables. To further improve TSO’s search efficiency and spatial exploration, this study introduces Sine–Logistic chaotic mapping, Levy flight, and Gaussian random walk strategy to optimize the TSO algorithm, developing the improved Tuna Swarm Optimization (ITSO). ITSO-optimized Stacking model achieved superior performance in DMI assessment, with an accuracy of 95.84%, significantly outperforming SVR, RF, DT, GBR, ETR, and AdaBoost models. This study provides a robust tool for precision feeding, contributing to optimizing cow feeding strategies, improving farm efficiency, and supporting sustainable dairy farming practices.

Accurate classification of urban tree species is fundamental for urban green space management and ecological assessment. To address the challenges of small and overlapping tree crown detection in high-resolution remote sensing imagery, this study proposes YOLO-CNGD, a novel framework based on YOLOv11n. The key enhancements include the integration of the Convolutional Block Attention Module (CBAM) for refined feature representation, the adoption of the Normalized Wasserstein Distance (NWD) loss for robust small-object localization, the incorporation of Deformable Convolution v3 (DCNv3) to adapt to irregular shapes, and the replacement of standard convolutions with GhostConv for a lightweight design. Experiments on a self-built urban tree dataset show that YOLO-CNGD achieves a precision of 94.8%, a recall of 91.1%, and an mAP@0.5 of 93.7%. The model balances accuracy and efficiency, showing great potential for large-scale automated urban tree inventory.

Dairy cow body condition score (BCS) is a practical, semi-quantitative indicator of body energy reserves and changes in energy balance. To improve five-class BCS detection under complex imaging conditions, this study developed UGT-YOLO by integrating a UniRepLKNet Block, a Gather-and-Distribute feature-fusion mechanism, and a Task-Aligned Dynamic Detection Head into YOLOv11n. The model was evaluated using a single publicly available dataset containing five adjacent BCS classes: 3.25, 3.50, 3.75, 4.00, and 4.25. Following redundancy removal and image-quality screening, 7015 original images were retained. Dataset partitioning was completed before data augmentation. The original images were divided at the source-video-sequence level into training, validation, and test subsets containing 5612, 702, and 701 images, respectively. Available cow identifiers were additionally used to keep images of the same identified animal within a single subset. Data augmentation was applied exclusively to the training subset, increasing the training set to 9639 images and producing a final experimental dataset of 11,042 images. UGT-YOLO achieved a precision of 84.3%, a recall of 81.1%, an mAP@0.5 of 88.5%, and an mAP@0.5:0.95 of 66.8%. Compared with YOLOv11n, these values increased by 4.8, 0.9, 2.8, and 4.5 percentage points, respectively. The parameter count increased from 2.6 to 6.5 million, and computational cost increased from 6.4 to 19.2 GFLOPs. Under an input resolution of 640 × 640 pixels, a batch size of 1, and FP32 inference on an NVIDIA GeForce RTX 3090, UGT-YOLO achieved a throughput of 68 frames s⁻¹. These results demonstrate an accuracy–complexity trade-off within the evaluated public dataset and restricted BCS range. Independent cow-level, cross-farm, full-range BCS, multi-scorer, and edge-device validation remains necessary.