The accuracy of discrete element simulations for silage corn stover is highly dependent on the precise calibration of model parameters. Addressing the relative scarcity of research on identifying DEM parameters for silage corn stover, this study constructs a simplified DEM model based on the Bonding constitutive model for granular materials. Parameter calibration is performed using experimental data on key physical and mechanical properties of the stover. Using the entire shear history stress-strain curve as the calibration benchmark, Gaussian process regression was introduced as a surrogate model. With mean squared error (MSE) as the objective function, a Bayesian optimization algorithm was employed to accurately identify the bonding parameters of the DEM model. The optimal parameter combination yielding the minimum MSE (MSE = 0.0072) was obtained: normal bonding stiffness, tangential bonding stiffness, normal strength, and shear strength were 5.12 & times; 109 N/m3, 1.28 & times; 108 N/m3, 1.60 & times; 107Pa, and 2.48 & times; 106 Pa, respectively. To validate this parameter set, three-point bending tests were conducted and compared with simulation results. The bending stress-strain curves from the discrete element model closely matched experimental trends and peak characteristics, confirming the model's accuracy. The proposed Bayesian optimization-based parameter calibration method demonstrates high precision and efficiency. It provides reliable references for discrete element simulations of silage corn stalk processing and the design optimization of key components in harvesting machinery.

In modern intensive pig farming, accurate and automated monitoring of feeding behavior is crucial for pig health assessment and production efficiency. A major challenge is the precise discrimination between "feeding" and "foraging" behaviors, which are highly similar in static images, while mainstream deep learning-based spatiotemporal models entail significant deployment costs and are heavily reliant on massive training datasets. We propose a novel method for pig feeding behavior recognition that combines lightweight object detection with spatiotemporal feature analysis. The method simplifies the complex behavior recognition task into a two-stage "object detection + spatiotemporal judgment" process. First, an optimized keypoint detection model (LMYOLO) is constructed to filter out non-feeding pigs by exclusively targeting "standing" individuals and their "head" keypoints. Second, a concise spatiotemporal discrimination algorithm is designed, which deconstructs feeding behavior into two computable physical conditions: the head being within the feeding area (a spatial constraint) and the body remaining relatively stationary for a preset duration (a temporal constraint). Experimental results show that the proposed method improves the F1f for feeding behavior recognition to 97.6 %, significantly outperforming a baseline model that relies solely on static information. For feeding time detection, the coefficient of determination R2 between the detected and ground-truth durations is 0.990, with a MAE of only 1.03 s. Furthermore, the complete algorithm has a size of only 4.85 MB, which is substantially smaller than mainstream spatiotemporal network models. This study provides a practical and cost-effective solution for implementing accurate pig feeding monitoring in cost-sensitive commercial farming environments.

In precision livestock farming, 3D point clouds provide important data support for analyzing pig behavior and monitoring their health. However, due to environmental occlusions, limited sensor viewpoints, and mutual shielding between pigs, the acquired point clouds are often severely partial, which affects the accuracy of body shape modeling and behavior recognition. To address these challenges, this study constructed a pig pose point cloud dataset using multi-view depth camera acquisition and point cloud registration techniques. Based on this dataset, an improved point cloud completion model, IIR-PoinTr, is proposed to enhance the reconstruction of geometric and topological structures in pig bodies. By strengthening local geometric perception and high-dimensional feature representation, the model improves the reconstruction quality of partial pig point clouds and produces more structurally consistent pig body shapes. Experimental results show that, on the self-constructed pig posture dataset, the proposed method reduces Chamfer Distance (CD-L1) by 3.6%, CD-L2 by 6.9%, and Earth Mover’s Distance (EMD) by 2.0%, while improving the F-score by 5.4% compared with the baseline model. In single-view point cloud completion tasks, the method is capable of reconstructing geometrically consistent pig body structures and increases downstream classification accuracy by 34.9%. These results indicate that the proposed method can improve the reconstruction quality of partial pig point clouds and provide preliminary technical support for posture analysis under occlusion.

Autonomous navigation for inspection robots in cattle barns critically depends on localization and path-planning algorithms. To address the issues of low navigation accuracy, long planning time, and insufficient trajectory smoothness in barn environments, this study proposes a navigation framework that combines LiDAR-inertial odometry and a rapidly-exploring random tree (LIO-RRTNav). For relocalization, this study develops a Fast_LIO2 with relocalization and pose optimization method (RP-Fast_LIO2). It exploits the geometric structure of the barn to associate the current frame point cloud with historical point clouds to suppress drift. Meanwhile, stable features are extracted from dynamic scenes and registered to the global map using the Iterative Closest Point (ICP) algorithm; the registration result is used as the initial state for an iterated extended Kalman filter (IEKF) to refine pose estimation. For path planning, this study proposes a highly efficient, robust, and smooth rapidly exploring random tree method (HRS-RRT). By incorporating a goal-oriented random sampling strategy and a safety-distance constraint, it enables safe and efficient planning, and the resulting path is further optimized through path pruning and cubic B-spline smoothing based on a spring potential energy model. Simulation test results showed that the HRS-RRT algorithm reduced 22.17%, 75.00%, and 83.09% in terms of path length, planning time and number of iterations, respectively, when compared with the traditional RRT algorithm. The experimental results revealed that RP-Fast_LIO2 algorithm reduced the mean and root mean square error of the absolute position error by 83.21% and 79.89%, when comparing with the traditional Fast_LIO2 algorithm. In navigation experiments conducted in two cattle yards, when the robot operated at speeds of 0.3, 0.5, and 1.0 m/s, the maximum lateral and longitudinal deviations did not exceed 0.13 and 0.08 m, respectively, and the maximum heading error did not exceed 7.24 degrees. The results acquired verified the adaptability of the LIO-RRTNav algorithm in the cattle yard environment, meeting the requirements of cattle yard inspection robot operation.

Mental and physical well-being is a prime factor that allows laying hens to exhibit natural behaviour. It is difficult to monitor the heat stress behavioral changes in laying hens for their welfare and production efficiency. This study proposed the real-time detection of those changes. However, three problems arise in battery-cage systems: birds occlude each other, behaviour labels are coarse, and edge devices have limited resources. To solve these problems, we proposed a three-dimensional spatiotemporal action detector named CSP-YOWO-TrajNet. A 2-D CNN and a 3-D CNN were designed to track and perceive trajectories to address these issues. Initially, CSPDarknet53-SPA was inserted into the 2-D branch to strengthen spatial feature extraction. Second, 3-D ResNeXt-50 was adopted in the 3-D branch for efficient temporal modelling. Finally, a TrajNet module was added to predict trajectories and improve perception and tracking. A video data set of 281 hen-behaviour clips was built for evaluation. The proposed CSP-YOWO-TrajNet achieved a precision of 94.1 % and an mAP@50 of 96.1 %. Compared with the YOWO baseline, precision and mAP@50 were raised by 3.0 % and 3.6 %, respectively, and the model size was reduced from 134.0 MB to 78.8 MB. In terms of tracking, the F1 score reached 95.4 % and the tracking accuracy reached 89.2 %. The proposed detector therefore supports real-time recognition of heat-stress behaviours in laying hens and can be deployed on edge devices.

Continuous identity output of individual dairy cows is a prerequisite for quantifying feeding duration, feeding frequency, and other behaviour-linked health indicators. In commercial feeding areas, long camera-to-animal distances, changing head posture, fence obstruction, neighbouring-animal occlusion, glare, and night-time infrared imaging frequently interrupt collar-digit visibility and make frame-by-frame recognition unreliable. This study presents Cattle-Vision, a vision-based system that combines a high-contrast bilaterally labelled collar, a complementary dual-camera layout, small-target head-and-neck detection, local optical character recognition, camera-specific multi-object tracking, and trajectory-level identity management. The system formulation distinguishes complementary dual-camera coverage from cross-camera tracking: Camera A and Camera B are processed independently, and identity continuity is evaluated within each camera view. For every matched detection-track pair, the head-and-neck region is cropped before OCR; the resulting digit sequence therefore inherits the corresponding camera-specific Track ID. A state-based identity cache performs initialization, short-term retention, historical voting, conflict handling, and post-occlusion re-confirmation. On the held-out detector test set, the full Hyper-FocusNet configuration achieved 99.7% precision, 95.1% recall, 96.2% mAP@50, and 89.3% mAP@50-95 at 38 detector-only FPS. On the independent continuous-video test set, Cattle-Vision achieved 91.32% individual-level identification accuracy, 88.67% IDF1, 88.72% identity-maintenance accuracy, and 92.13% re-confirmation success. The results demonstrate that track-conditioned OCR and trajectory-level identity management reduce unknown outputs, identity interruptions, and erroneous identity switches caused by temporary collar occlusion, OCR failure, or missed detections.

Mulberry (Morus spp.) leaves, fruits, branches, and root bark are rich in bioactive compounds, including 1-deoxynojirimycin (1-DNJ) and γ-aminobutyric acid (GABA), supporting their potential use in food, medicinal, and feed applications. Their high moisture content, however, makes fresh materials highly susceptible to postharvest quality deterioration, making drying essential for stabilization and high-value utilization. Drying technologies involve trade-offs among efficiency, energy consumption, sensory quality, rehydration, and bioactive-compound retention. This review provides a comprehensive overview of pretreatment and drying technologies for mulberry materials, with particular attention to differences in raw-material characteristics, processing conditions, analytical methods, and reporting bases that limit direct comparisons among studies. Current evidence suggests that low-temperature, low-oxygen, or short-duration technologies, including vacuum freeze-drying, microwave drying, and microwave-vacuum drying, may better preserve quality in certain thermosensitive products, although their benefits remain product- and process-dependent. Hot-air, solar, infrared, heat-pump, and hybrid drying remain practical options for bulk products but require optimization to balance quality, energy efficiency, and scalability. For juice and functional powders, carrier selection, powder properties, and bioaccessibility require further study. Overall, the available evidence is heterogeneous, and some conclusions rely on limited mulberry-specific data or extrapolation from related plant matrices. Future research should emphasize standardized quality evaluation, harmonized reporting, kinetic modeling, multi-objective optimization, online monitoring, energy and carbon-footprint assessment, and industrial-scale validation.

To enhance agricultural machinery operations' stability, reliability, and accuracy in complex field uneven, this paper presents the development of a self-driven, finger-inserted rotational triboelectric angle sensor (SFRTAS). This sensor uses triboelectric nanogenerator (TENG) and flexible printed circuit board (FPCB) technology. When connected to a ground contour-following curved rod, the SFRTAS forms a self-driven triboelectric synchronized contour-following detection system. As the contour-following curved rod moves along uneven ground, it follows surface undulations, driving the sensor to rotate and generate continuous electrical signals containing angular information, thereby achieving ground contour-following. The developed SFRTAS has a diameter of approximately 40 mm, a thickness of 10.19 mm, and a weight of 6.58 g. It incorporates three sets of triboelectric electrode array structures with phase differences, achieving a sensing resolution of 0.188 mm, a sensitivity of 7.69 P & sdot;mm-1 , an angular precision of 0.5 degrees, durability exceeding 1 million cycles, low hysteresis, and strong resistance to environmental interference. Digital soil-trough experiments were conducted on the contour-following detection system to validate its performance. The results demonstrated that the system could continuously measure ground uneven information during operation. Neither the forward speed of the test vehicle nor changes in ridge height affected the real-time performance of the sensor signals. The system exhibited high measurement accuracy, excellent sensitivity, and strong reliability. Moreover, it can accurately, rapidly, and continuously detect ground uneven variations, making it highly applicable to the field of agricultural machinery.

This study introduces a portable radar system for real-time monitoring of respiratory and heart rates in dairy cows. It uses millimeter-wave Frequency Modulated Continuous Wave (FMCW) radar to perform noncontact physiological sensing, reducing behavioral disturbance. The radar system's design prioritizes portability, cost-effectiveness, and robustness, allowing deployment in diverse farm environments. Experimental results show strong agreement between radar-derived and reference measurements, confirmed through correlation analysis and supervised classification. Additionally, the study explores the integration of radar monitoring with advanced data analysis techniques, including Principal Component Analysis (PCA) and Support Vector Machines (SVM), to enhance livestock health management processes. The system is validated for its capability to monitor respiratory and heart rates in real time and effectively classify cows' reproductive states, achieving a classification accuracy of 79.63% for estrus detection. These findings demonstrate the feasibility of radar-based physiological monitoring and support future integration with data-driven management tools. IEEJ Transactions on Electrical and Electronic Engineering published by Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.

This study employs multi-sensor data fusion, signal analysis, and machine learning techniques to monitor and identify estrus-specific behaviors, such as frequent circling and restless standing in dairy cows, to enhance the accuracy and efficiency of estrus detection. The fast Fourier transform (FFT) was applied to time series data to identify specific frequency patterns associated with estrous behaviors. Principal component analysis (PCA) was used for dimensional reduction of behavioral data, effectively revealing complex patterns and structures within the data, allowing clear differentiation between estrous and non-estrous behaviors. The results demonstrate that the adoption of advanced technologies and algorithms significantly improves the performance of estrus behavior monitoring systems.

Stubble cutting is a critical step in the mechanized harvesting of mulberry trees. Poor stubble cutting quality can lead to root decay, and even the death of the trees, reducing the rate of rejuvenation. In view of the relatively high damage rate of stubble cutting in the mulberry harvesting operations, this paper has conducted in-depth research and designed a mulberry branch harvesting and stubble cutting test machine. The overall structure and key technical parameters of the machine were determined and the design of critical components was theoretically analyzed. Field tests identified an optimized set of cutting parameters that effectively reduced stubble damage and energy consumption. The optimal parameter combination included a saw blade line speed of 74 m/s, a star-wheel ground clearance of 727 mm, and a cutting speed ratio of 93, resulting in a cutting energy consumption within 1 m of 241 mJ and a stubble score of 8.5. These parameters met the quality standards for mulberry harvesting. This research provides valuable data support for analyzing and optimizing the cutting parameters of mulberry stubble-cutting machines, laying a solid foundation for advancing mechanized harvesting technology for mulberry trees.

Accurate and non-destructive detection of total nitrogen (TN), total phosphorus (TP), and total potassium (TK) levels in soil is crucial for precise soil testing and fertilization in modernized precision agriculture. Traditional methods for soil composition analysis are expensive, time-consuming, and destructive. This research aims to establish a low-cost, high-precision, and non-destructive method for soil nutrient detection based on visible-near-infrared (Vis-NIR) spectroscopy (350-2500 nm) combined with improved machine learning algorithms. The VisNIR spectra of soil samples were acquired using the RS-5400 high-resolution ground feature spectrometer. Subsequently, the Monte Carlo sampling cross-validation (MCCV) algorithm was used to eliminate abnormal samples, and then different preprocessing methods were performed on the spectral data including first-derivative (FD), Savitzky-Golay smoothing (SG) and others. The optimal preprocessing method was selected from these options. In order to remove redundant information and increase the speed of calculation, five algorithms such as competitive adaptive reweighted sampling (CARS), iteratively retains informative variables (IRIV) and the variable iterative space shrinkage approach (VISSA)-IRIV algorithm were used to select feature variables. The characteristic wavelengths closely related to TN, TP, and TK in the soil have been extracted. Then, the RBF kernel (radial basis function) and poly kernel were mixed to obtain the RBF-poly hybrid kernel function, and then the hybrid kernel function support vector machine (RBF-poly-SVM) and the radial basis kernel function support vector machine (RBF-SVM) were applied respectively. Establish prediction models and introduce the whale optimization algorithm (WOA) to optimize the g (kernel function parameter), c (penalty factor) and k-rbf (weight coefficient) parameters in the two models. The performance of the developed models was tested using the coefficient of determination (R2), the root mean squared error (RMSE) and the ratio of performance to deviation (RPD). The results demonstrated that among all models, the RBF-poly-SVM modeling methods were superior to the RBF-SVM model. The best results for estimation of TN, TP, and TK elements were achieved by the models of SG-square-FD + IRIV + RBF-poly-SVM (R2C=0.960, R2V=0.902, RPD=3.206), square-FD + IRIV + RBF-poly-SVM (R2C=0.999, R2V=0.937, RPD=3.939), square root + VISSA-IRIV + RBF-poly-SVM (R2C=0.955, R2V=0.904, RPD=2.608), respectively. The findings of the current approach own practical implications for agriculture and environmental management, as they enable more efficient and accurate soil nutrient monitoring and management.

The effect of high-voltage electrostatic field (HVEF) on maize seeds’ resistance to chilling injury remains unclear. This study investigates the chemical and spatial changes induced by HVEF at macroscopic and microscopic levels via the combination of physiological assessments and scanning electron microscopy (SEM). Maize samples were categorized into low and normal-temperature groups. At an HVEF strength of 1.6 kV/cm, all indices in the low-temperature group significantly improved relative to the control (P < 0.01), with germination potential, rate, index, and vigor index increasing by 11.7%, 11.2%, 10.5%, and 31.7%, respectively. Root length, shoot length, and dry weight of maize seedlings rose by 20.3%, 19.2%, and 16.6%. Further analysis revealed a 62.7% increase in soluble sugar content in HVEF-treated seeds and the lowest leaching solution conductivity of 1.6 kV/cm. These results demonstrate that HVEF treatment enhances soluble sugar accumulation during seed germination, regulating osmotic balance within the cells. Furthermore, SEM assessed maize microtissue morphology after chilling injury and HVEF treatment. Optimal HVEF treatment resulted in cell wall expansion, enhanced fiber elasticity, reduced interstitial spaces, and swollen cells, indicating improved hydrophilicity and protease activity. This study offers valuable insights into the mechanisms by which HVEF improves seed performance under low-temperature stress.

Weed growth significantly impacts corn yield. With the continuous development of weed control technologies, achieving more effective and precise weed management has become a major challenge in corn production. To achieve precise weed suppression, this study proposes a growth point detection method based on a keypoint pose estimation model capable of effectively detecting various weeds and locating various weed growth points during the 2nd-5th leaf stage of corn development. To address the complex working environment of precision weeding machines in corn fields, including occlusion, dense growth, and variable lighting conditions, we design a dilation-wise residual module (DWRM) for the detector and a separation and enhancement attention module (SEAM) for pose estimation to adapt to these challenges. Furthermore, owing to the limited computational resources in field settings, we introduced the RepViT block (RVB) to achieve model lightweighting. The proposed method was evaluated on the constructed corn field dataset. The experimental results demonstrated that SRD-YOLO achieved an m A P k p t of 96.5 ​%, an F1 score of 94 ​%, and an FPS of 169, while reducing the model parameters by 8.7M. SRD-YOLO effectively meets the requirements for growth point localization under challenging conditions, providing robust technical support for real-time and precise weed control in corn fields.

The co-rearing model for young silkworms (Bombyx mori) utilizing artificial feed is currently undergoing significant promotion within the sericulture industry in China. Within this model, accurately counting the number of young silkworms serves as a crucial foundation for achieving precision rearing and high-quality breeding. Currently, manual counting remains the prevalent method for enumerating young silkworms, yet it is highly subjective. A dataset of young silkworm bodies has been constructed, and the Young Silkworm Counting (YSC) method has been proposed. This method combines an improved detector, incorporating an optimized multi-scale feature fusion module and the Efficient Multi-Scale Attention Fusion Cross Stage Partial (EMA-CSP) mechanism, with an optimized tracker (based on ByteTrack with improved detection box matching), alongside the implementation of a ‘detection line’ approach. The experimental results demonstrate that the recall, precision, and average precision (AP50:95) of the improved detection algorithm are 87.9%, 91.3% and 72.7%, respectively. Additionally, the enhanced ByteTrack method attains a multiple-object tracking accuracy (MOTA) of 88.3%, an IDF1 of 90.2%, and a higher-order tracking accuracy (HOTA) of 78.1%. Experimental validation demonstrates a counting accuracy exceeding 90%. The present study achieves precise counting of young silkworms in complex environments through an improved detection-tracking method combined with a detection line approach.

Accurate and efficient detection of tachypnea and salivation behavior plays a key role in improving the health management of beef cattle. To address the challenges of low detection accuracy and high computational redundancy in existing algorithms within complex breeding environments, the Cattle-ES3D algorithm was proposed for detecting tachypnea and salivation behaviors in beef cattle. First, a hybrid architecture was proposed, which integrated the Embedded Spatial Pyramid Network (ESP-Net) for multi-scale extraction and SlowFast dual-pathway network to realize the extraction of spatiotemporal features in beef cattle. Second, the Adaptive Spatiotemporal Feature Fusion Synchronization Module (AST-Sync) was designed to achieve adaptive fusion of spatiotemporal features. Finally, a lightweight dynamic detection branch was designed to achieve classification-regression feature spatial alignment and temporal association constraints through a multi-dimensional parameter optimization mechanism driven by spatiotemporal dynamic label assignment. The experimental results showed that the method utilized resulted with spatial and temporal features to enhance the detection accuracy of tachypnea and salivation behaviors in beef cattle, with reduced computational cost. The Cattle-ES3D model achieved a mean Average Precision (mAP) of 93.4 %, GFLOPs of 39.6 and FPS of 33.2. Compared to C3D, I3D, P3D and R(2 + 1)D, Cattle-ES3D improved mAP by 8.2 %, 6.8 %, 3.3 %, and 18.4 % respectively, while reducing GFLOPs by 0.7, 11.8, 7.0, and 8.2 respectively. These results demonstrated that the proposed model provided a robust and high-performance technical solution for intelligent livestock farming.

The current study was conducted to address the challenges of recognizing alfalfa seed pod maturity in complex field environments, and the significant impact of the quantity of labeled samples on the performance of object detection algorithms. A method for identifying the maturity of alfalfa seed pod clusters was proposed using an unmanned aerial vehicle (UAV) and a semi-supervised deep learning model SSOD-MViT (Semi-Supervised Object Detection based on the MViTNet). To enhance the model's capability to extract key feature information, an improved lightweight general vision transformer MobileViT (Mobile Vision Transformer) was firstly employed as the backbone. The deep integration of ScConv (Spatial and Channel Reconstruction Convolution) was additionally employed to reduce redundant information within the channels, thereby decreasing the computational load of the model. Secondly, a small object detection layer was incorporated into the Neck, and the Efficient Multi-Scale Attention Module (EMA) was added to the C2f structure. The SAHI (Slicing Aided Hyper Inference) algorithm was integrated during the inference process, which improves the detection accuracy of small-sized alfalfa seed pod clusters and enhances the model's resistance to interference. Finally, the concept of Consistency Regularization was incorporated into the model to reduce its dependency on sample data. The experimental results revealed that SSOD-MViT achieved a mAP@0.5 of 92.23 %. When compared to the YOLOv8 object detection model, the mAP@0.5 had improved by 12.31 %. When compared to the Faster R-CNN object detection model, the average detection time reduced by 175.81 ms. The proposed model MViTNet (MobileViT Network) had a storage size of 5.3 MB, and an average detection time of 82.34 ms, providing favorable conditions for subsequent deployment on embedded devices. This research effectively improved the detection performance of existing models in detecting alfalfa seed pod maturity in complex field environments. This advancement also aids in determining the optimal harvesting period for alfalfa seeds, thereby providing technical support to enhance productivity and reduce production costs in the alfalfa seed production industry.

Enhancing the growth rate and shortening the growth cycle of mung bean sprouts while maintaining quality remains a significant challenge. This study presents a novel approach using pulsed electric field (PEF) treatment to improve both the quality and absolute growth rate of mung bean sprouts. Specifically, mung bean seeds were pretreated using PEF, and the effects of PEF intensity and frequency on the growth of mung bean sprouts were analyzed based on orthogonal germination experiments. Optimal treatment conditions were identified as a PEF intensity of 119.2 kV/m, a frequency of 9.0 Hz, and a treatment duration of 100 s. Under these conditions, mung bean sprouts exhibited substantial increases in total length, hypocotyl length, and fresh weight, with fresh weight rising by 22.3 % compared to the control group (P < 0.01). Nutrient analysis revealed significant increases in protein (28.1 %), carbohydrate (21 %), and ash content (14.3 %) post-treatment, highlighting PEF's effectiveness in enhancing the nutritional profile of mung bean sprouts. Conversely, we noted the potential risks of excessive PEF intensity, which could lead to irreversible cellular damage. The PEF method proved to be a rapid and effective technique for boosting mung bean sprout growth, shortening growth cycle, and enhancing sprout yield.

The classification of silkworm cocoons is essential prior to silk reeling and serves as a key step in improving the quality of raw silk. At present, cocoon classification mainly relies on manual sorting, which is labor-intensive and inefficient. In this paper, a cocoon detection algorithm S-YOLOv8_c based on the cooperation of MobileSAM and YOLOv8 for the mountage cocoons was proposed. The MobileSAM with a designed area thresholding algorithm was used for the semantic segmentation of mountage cocoon images, which could mitigate the effect of complex backgrounds and maximize the discriminability of cocoon features. Subsequently, the BiFPN was added to the neck of YOLOv8 to improve the multiscale feature fusion capability. The loss function was replaced with the WIoU, and a dynamic non-monotonic focusing mechanism was introduced to improve the generalization ability. In addition, the GAM was incorporated into the head to focus on detailed cocoon information. Finally, the S-YOLOv8_c achieved a good detection accuracy on the test set, with a mAP of 95.8%. Furthermore, to experimentally validate the sorting ability, we deployed the proposed model onto the self-developed Cartesian coordinate automatic cocoon harvester, which indicated that it would effectively meet the requirements of accurate and efficient cocoon sorting.

The enclosed multi-story poultry housing is a type of poultry enclosure widely used in industrial caged chicken breeding. Accurate identification and detection of the comb and eyes of caged chickens in poultry farms using this type of enclosure can enhance managers’ understanding of the health of caged chickens. However, the accuracy of image detection of caged chickens will be affected by the enclosure's entrance, which will reduce the precision. Therefore, this paper proposes a cage-gate removal algorithm based on big data and deep learning Cyclic Consistent Migration Neural Network (CCMNN). The method achieves automatic elimination and restoration of some key information in the image through the CCMNN network. The Structural Similarity Index Measure (SSIM) between the recovered and original images on the test set is 91.14%. Peak Signal-to-Noise Ratio (PSNR) is 25.34dB. To verify the practicability of the proposed method, the performance of the target detection algorithm is analyzed both before and after applying the CCMNN network in detecting the combs and eyes of caged chickens. Different YOLOv8 detection algorithms, including YOLOv8s, YOLOv8n, YOLOv8m, and YOLOv8x, were used to verify the algorithm proposed in this paper. The experimental results demonstrate that compared to images without CCMNN processing, the precision of comb detection of caged chickens is improved by 11%, 11.3%, 12.8%, and 10.2%. Similarly, the precision of eye detection for caged chickens is improved by 2.4%, 10.2%, 6.8% and 9%. Therefore, more complete outline images of caged chickens can be obtained using this algorithm and the precision in detecting the comb and eyes of caged chickens can be enhanced. These advancements in the algorithm offer valuable insights for future poultry researchers aiming to deploy enhanced detection equipment, thereby contributing to the accurate assessment of poultry production and farm conditions.