There are some problems with estrus detection in ewes in large-scale meat sheep farming: mainly, the manual detection method is labor-intensive and the contact sensor detection method causes stress reactions in ewes. To solve the abovementioned problems, we proposed a multi-objective detection layer neural network-based method for ewe estrus crawling behavior recognition. The approach we proposed has four main parts. Firstly, to address the problem of mismatch between our constructed ewe estrus dataset and the YOLO v3 anchor box size, we propose to obtain a new anchor box size by clustering the ewe estrus dataset using the K-means++ algorithm. Secondly, to address the problem of low model recognition precision caused by small imaging of distant ewes in the dataset, we added a 104 × 104 target detection layer, making the total target detection layer reach four layers, strengthening the model’s ability to learn shallow information and improving the model’s ability to detect small targets. Then, we added residual units to the residual structure of the model, so that the deep feature information of the model is not easily lost and further fused with the shallow feature information to speed up the training of the model. Finally, we maintain the aspect ratio of the images in the data-loading module of the model to reduce the distortion of the image information and increase the precision of the model. The experimental results show that our proposed model has 98.56% recognition precision, while recall was 98.04%, F1 value was 98%, mAP was 99.78%, FPS was 41 f/s, and model size was 276 M, which can meet the accurate and real-time recognition of ewe estrus behavior in large-scale meat sheep farming.

We propose a lightweight neural network-based method to detect the estrus behavior of ewes. Our suggested method is mainly proposed to solve the problem of not being able to detect ewe estrus behavior in a timely and accurate manner in large-scale meat sheep farms. The three main steps of our proposed methodology include constructing the dataset, improving the network structure, and detecting the ewe estrus behavior based on the lightweight network. First, the dataset was constructed by capturing images from videos with estrus crawling behavior, and the data enhancement was performed to improve the generalization ability of the model at first. Second, the original Darknet-53 was replaced with the EfficientNet-B0 for feature extraction in YOLO V3 neural network to make the model lightweight and the deployment easier, thus shortening the detection time. In order to further obtain a higher accuracy of detecting the ewe estrus behavior, we joined the feature layers to the SENet attention module. Finally, the comparative results demonstrated that the proposed method had higher detection accuracy and FPS, as well as a smaller model size than the YOLO V3. The precision of the proposed scheme was 99.44%, recall was 95.54%, F1 value was 97%, AP was 99.78%, FPS was 48.39 f/s, and Model Size was 40.6 MB. This study thus provides an accurate, efficient, and lightweight detection method for the ewe estrus behavior in large-scale mutton sheep breeding.

In order to solve the problems of low efficiency and subjectivity of manual observation in the process of group-sheep-aggression detection, we propose a video streaming-based model for detecting aggressive behavior in group sheep. In the experiment, we collected videos of the sheep’s daily routine and videos of the aggressive behavior of sheep in the sheep pen. Using the open-source software LabelImg, we labeled the data with bounding boxes. Firstly, the YOLOv5 detects all sheep in each frame of the video and outputs the coordinates information. Secondly, we sort the sheep’s coordinates using a sheep tracking heuristic proposed in this paper. Finally, the sorted data are fed into an LSTM framework to predict the occurrence of aggression. To optimize the model’s parameters, we analyze the confidence, batch size and skipping frame. The best-performing model from our experiments has 93.38% Precision and 91.86% Recall. Additionally, we compare our video streaming-based model with image-based models for detecting aggression in group sheep. In sheep aggression, the video stream detection model can solve the false detection phenomenon caused by head impact feature occlusion of aggressive sheep in the image detection model.

In large-scale meat sheep farming, high CO₂ concentrations in sheep sheds can lead to stress and harm the healthy growth of meat sheep, so a timely and accurate understanding of the trend of CO₂ concentration and early regulation are essential to ensure the environmental safety of sheep sheds and the welfare of meat sheep. In order to accurately understand and regulate CO₂ concentrations in sheep barns, we propose a prediction method based on the RF-PSO-LSTM model. The approach we propose has four main parts. First, to address the problems of data packet loss, distortion, singular values, and differences in the magnitude of the ambient air quality data collected from sheep sheds, we performed data preprocessing using mean smoothing, linear interpolation, and data normalization. Second, to address the problems of many types of ambient air quality parameters in sheep barns and possible redundancy or overlapping information, we used a random forests algorithm (RF) to screen and rank the features affecting CO₂ mass concentration and selected the top four features (light intensity, air relative humidity, air temperature, and PM2.5 mass concentration) as the input of the model to eliminate redundant information among the variables. Then, to address the problem of manually debugging the hyperparameters of the long short-term memory model (LSTM), which is time consuming and labor intensive, as well as potentially subjective, we used a particle swarm optimization (PSO) algorithm to obtain the optimal combination of parameters, avoiding the disadvantages of selecting hyperparameters based on subjective experience. Finally, we trained the LSTM model using the optimized parameters obtained by the PSO algorithm to obtain the proposed model in this paper. The experimental results show that our proposed model has a root mean square error (RMSE) of 75.422 μg·m⁻³, a mean absolute error (MAE) of 51.839 μg·m⁻³, and a coefficient of determination (R²) of 0.992. The model prediction curve is close to the real curve and has a good prediction effect, which can be useful for the accurate prediction and regulation of CO₂ concentration in sheep barns in large-scale meat sheep farming.

Sheep aggression detection is crucial for maintaining the welfare of a large-scale sheep breeding environment. Currently, animal aggression is predominantly detected using image and video detection methods. However, there is a lack of lightweight network models available for detecting aggressive behavior among groups of sheep. Therefore, this paper proposes a model for image detection of aggression behavior in group sheep. The proposed model utilizes the GhostNet network as its feature extraction network, incorporating the PWConv and Channel Shuffle operations into the GhostConv module. These additional modules improve the exchange of information between different feature maps. An ablation experiment was conducted to compare the detection effectiveness of the two modules in different positions. For increasing the amount of information in feature maps of the GhostBottleneck module, we applied the Inverted-GhostBottleneck module, which introduces inverted residual structure based on GhostBottleneck. The improved GhostNet lightweight feature extraction network achieves 94.7% Precision and 90.7% Recall, and its model size is only 62.7% of YOLOv5. Our improved model surpasses the original model in performance. Furthermore, it addresses the limitation of the video detection model, which was unable to accurately locate aggressive sheep. In real-time, our improved model successfully detects aggressive behavior among group sheep.

Aiming at the challenges in practical production lines, including the difficulty in accurately capturing external defects on continuously rolling walnuts, distinguishing subtle defects, and differentiating narrow fissures from natural walnut textures, this paper proposes an improved walnut external defect detection model named YOLO11-GME, based on YOLO11n. Firstly, the original backbone network is replaced with the lightweight GhostNetV1 network, enhancing model precision while meeting real-time detection speed requirements. Secondly, a Mixed Local Channel Attention (MLCA) mechanism is incorporated into the neck to strengthen the network’s ability to capture features of subtle defects, thereby improving defect recognition accuracy. Finally, the EIoU loss function is adopted to enhance the model’s localization capability for irregularly shaped defects and reduce false detection rates by improving the scale sensitivity of bounding box regression. Experimental results demonstrate that the improved YOLO11-GME model achieves a mean Average Precision (mAP) of 96.2%, representing improvements of 8.6%, 7%, and 5.8% compared to YOLOv5n, YOLOv8n, and YOLOv10n, respectively, and a 5.9% improvement over the original YOLOv11. Precision rates for the normal, fissure, and inferior categories increased by 8.7%, 5.3%, and 3.7%, respectively. The frame rate remains at 43.92 FPS, approaching the original model’s 51.02 FPS. These results validate that the YOLO11-GME model enhances walnut external defect detection accuracy while maintaining real-time detection speed, providing robust technical support for defect detection and classification in industrial walnut production.