Computers and Electronics in Agriculture最新文献

{"title":"Dynamic behaviors of fertilizer droplets impacting tea leaf surfaces","authors":"Bin Hu ,&nbsp;Hong Li ,&nbsp;Yue Jiang ,&nbsp;Longfei Du","doi":"10.1016/j.compag.2025.110945","DOIUrl":"10.1016/j.compag.2025.110945","url":null,"abstract":"<div><div>With the emergence of dissolved organic fertilizers and new foliar fertilizers, the adoption of fertigation via sprinkler systems has become increasingly prevalent in tea cultivation across various provinces in China. However, extensive water and fertilizer management practices have hindered the enhancement of tea yield, quality, and economic benefits. The appropriate management mode is based on maximizing spray retention, which is significantly influenced by the impingement dynamics of fertilizer droplets on tea leaves. Nonetheless, research in this area is still relatively limited. In this study, the dynamic behaviors of fertilizer droplets impacting tea leaf surfaces (<em>Wu-Niuzao</em>) are explored through both experiments and a Coupled Level Set and Volume of Fluid (CLSVOF) interface-capturing method. The effects of Weber number and inclination angle <em>α</em> on droplet impact dynamics, including impact phenomenology, maximum spreading diameter, maximum spreading time and the associated principles, are thoroughly explored. The simulation predictions provide a well match with the experiment results, suggesting that the CLSVOF interface tracking approach can offer theoretical support for enhancing fertilization efficiency. Three distinct impact behaviors (deposition, receding breakup, and sliding-off) were systematically identified across ranges of Weber numbers and <em>α</em> values, which significantly influence the dynamics of drop impact. The findings indicate that as Weber number and <em>α</em> increase, the tangential component (<em>We</em>_T) and gravity forces are strengthened to promote the later droplet spread, result in additional droplet spreading, and lead to rivulet formation. The established quantitative relationships and corresponding results can assist in the rational selection and design parameters of sprinkler irrigation systems, providing valuable insights for enhancing fertilization efficiency in tea plantations in China.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"239 ","pages":"Article 110945"},"PeriodicalIF":8.9,"publicationDate":"2025-09-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145003579","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A top-down deep neural network for multi-dairy cows pose estimation and lameness detection","authors":"Saisai Wu ,&nbsp;Shuqing Han ,&nbsp;Xiaoxiang Mo ,&nbsp;Yingying Wei ,&nbsp;Yuanyuan Qin ,&nbsp;He Chen ,&nbsp;Jianzhai Wu ,&nbsp;Zhikang Zeng","doi":"10.1016/j.compag.2025.110911","DOIUrl":"10.1016/j.compag.2025.110911","url":null,"abstract":"<div><div>Cow pose estimation and real-time health monitoring are important for refined herd management, improved animal welfare, and reduced passive culling rates. However, existing multi-object pose estimation methods often struggle to adapt to multi-scale objects in complex environments and typically exhibit low accuracy in detecting occluded keypoints. To address these challenges, this study proposes a top-down deep neural network for multi-dairy cows pose estimation and lameness detection, which integrates lightweight object detection, multi-scale feature fusion, and comprehensive motion feature analysis to improve the robustness under complex farm conditions. First, the real-time object detector YOLOv8n is improved by introducing the Partial Convolution (PConv) and Slim-neck modules, which improve both the efficiency and accuracy of object bounding box predictions, providing a solid foundation for the subsequent pose estimation. Second, a Path Aggregation Feature Pyramid Network (PAFPN)-based multi-scale feature fusion module is introduced as the neck network within the Real-time Multi-person Pose Estimation (RTMPose). This is further supported by a transfer learning strategy to improve keypoint localization, particularly under-occlusion and scale variation conditions. The experimental results show that the improved model achieves a mean average precision (<em>mAP)</em> of 95.8 %, significantly outperforming the baseline model and other existing algorithms. Seven motion features, including gait symmetry, head swing amplitude, and back curvature, were extracted in real time through pose tracking and motion trajectory analysis. These features were normalized and input into a Random Forest classifier for lameness detection. The model was evaluated on a dataset of 418 dairy cows and achieved average accuracy, sensitivity, and specificity values of 93.8 %, 94.4 %, and 97.5 %, respectively. These results demonstrate that combining multiple motion features provides a more accurate assessment of lameness.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"239 ","pages":"Article 110911"},"PeriodicalIF":8.9,"publicationDate":"2025-09-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144997186","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Can robots “Taste” grapes? Estimating SSC with simple RGB sensors","authors":"Thomas A. Ciarfuglia ,&nbsp;Ionut M. Motoi ,&nbsp;Leonardo Saraceni ,&nbsp;Daniele Nardi","doi":"10.1016/j.compag.2025.110845","DOIUrl":"10.1016/j.compag.2025.110845","url":null,"abstract":"<div><div>In table grape cultivation, harvesting depends on accurately assessing fruit quality. While some characteristics, like color, are visible, others, such as Soluble Solid Content (SSC), or sugar content measured in degrees Brix (°Brix), require specific tools. SSC is a key quality factor that correlates with ripeness, but lacks a direct causal relationship with color. Hyperspectral cameras can estimate SSC with high accuracy under controlled laboratory conditions, but their practicality in field environments is limited. This study investigates the potential of simple RGB sensors under uncontrolled lighting to estimate SSC and color, enabling cost-effective, robot-assisted harvesting. Over the 2021 and 2022 summer seasons, we collected grape images with corresponding SSC and color labels to evaluate algorithmic solutions for SSC estimation, specifically testing for cross-seasonal and cross-device robustness. We propose two approaches: a computationally efficient histogram-based method for resource-constrained robots and a Deep Neural Network (DNN) model for more complex applications. Our results demonstrate high performance, with the DNN model achieving a Mean Absolute Error (MAE) as low as 1.05 °Brix on a challenging cross-device test set. The lightweight histogram-based method also proved effective, reaching an MAE of 1.46 °Brix. These results are highly competitive with those from hyperspectral systems, which report errors in the 1.27–2.20 °Brix range in similar field applications.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"239 ","pages":"Article 110845"},"PeriodicalIF":8.9,"publicationDate":"2025-09-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144988821","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Digital mapping of DTPA extractable micronutrients using combined AVIRIS-NG hyperspectral and multispectral data of Sentinel-2 and SRTM-DEM","authors":"Bhabani Prasad Mondal ,&nbsp;Rabi Narayan Sahoo ,&nbsp;Bappa Das ,&nbsp;Nayan Ahmed ,&nbsp;Kali Kinkar Bandyopadhyay ,&nbsp;Joydeep Mukherjee ,&nbsp;Alka Arora ,&nbsp;Ali Refaat Ali Moursy","doi":"10.1016/j.compag.2025.110905","DOIUrl":"10.1016/j.compag.2025.110905","url":null,"abstract":"<div><div>Real-time assessment and prediction of soil nutrients are essential for precision soil nutrient management. Digital mapping of soil nutrients, influenced by topographical attributes, offers a promising approach for site-specific nutrient management. Although, multispectral and hyperspectral remote sensing data have been utilized separately for the digital mapping of soil nutrients, the integration of both data types for soil nutrients especially for soil micronutrients remains unexplored. Therefore, this study attempted to enhance the accuracy and reliability of predicting four important Diethylene Triamine Pentaacetic Acid (DTPA) extractable micronutrient cations such as zinc (Zn), copper (Cu), iron (Fe), and manganese (Mn) in the Katol block of Nagpur district, Maharashtra, India by integrating multispectral data of Sentinel-2, and hyperspectral data of Airborne Visible Infrared Imaging Spectrometer-Next Generation (AVIRIS-NG) along with the Shuttle Radar Topographic Mission (SRTM) digital elevation model (DEM) data, soil and climatic data with three machine learning (ML) models. These four micronutrients were selected for digital mapping owing to their crucial roles in plant growth and development, and their continuous depletion under intensive cultivation of high-yielding varieties, which may adversely affect crop productivity and soil health in the study area. 132 georeferenced surface soil samples were collected during airborne hyperspectral spectra acquisition by the AVIRIS-NG. Micronutrients were analysed in laboratory and lab-generated data combined with multispectral, hyperspectral, soil and climatic data with the help of three ML models viz. Random Forest (RF), Quantile Random Forest (QRF) and Cubist for predicting and mapping of those four key micronutrients. Based on the prediction performances especially in the validation dataset, the RF model was found more promising in comparison to other models in micronutrient prediction. Since the RF model demonstrated superior performance, it was utilized for mapping those four micronutrients. This RF model demonstrated moderate to satisfactory accuracy, with the value of the coefficient of determination (R²) 0.71, 0.65, 0.62 and 0.63 for Zn, Cu, Fe, and Mn prediction respectively. The study utilized eight different types of datasets and applied them to each model and it was revealed that the combined use of multispectral, hyperspectral, soil and climatic datasets improved the prediction accuracy of micronutrients compared to using each dataset individually. The study also revealed that soil variable i.e., soil pH was the most influential variable for soil micronutrient especially for Zn, Cu and Fe predictions. Next to this, hyperspectral variables (PC2 for Zn, Fe, PC1, and PC3 for Mn, and PC6, and PC7 for Cu) were found crucial in predicting and mapping soil-available micronutrients. Furthermore, it was observed that after soil and hyperspectral variables, DEM-derived var","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"239 ","pages":"Article 110905"},"PeriodicalIF":8.9,"publicationDate":"2025-09-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144988812","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A validated low-cost model for precise water leak detection in irrigation ponds (WLDIP)","authors":"A.J. Zapata-Sierra,&nbsp;F. Manzano-Agugliaro","doi":"10.1016/j.compag.2025.110954","DOIUrl":"10.1016/j.compag.2025.110954","url":null,"abstract":"<div><div>Efficient water management in agriculture is crucial for global food security and environmental sustainability, especially in water-scarce regions. Undetected leaks in irrigation ponds lead to significant water losses and increased operational costs. This study aimed to propose a low-cost Water Leak Detection model for Irrigation Ponds (WLDIP), capable of estimating both the location (height) and size of leaks, thereby helping to reduce agricultural water losses. The methodology integrates continuous water level measurements with a detailed water balance model, considering variables such as inflows, outflows, evaporation, and thermal expansion of both the water and the reservoir structure. The model relies on measuring water height every 15 min over a week and uses meteorological data to account for evaporation and thermal expansion. Least squares optimization compares measured data with the model’s calculated values to determine leak parameters. Field validation was conducted on three irrigation ponds in southeastern Spain. The WLDIP model successfully identified leak locations and sizes in all three ponds, which were subsequently confirmed by physical inspections and repairs. Results showed that level variations exceeding 0.01 m within 15 min were due to normal pond use, while leaks exhibited a typical flow rate ten times lower. This demonstrates the model’s high accuracy in determining both the height (h_LEAK) and size (ω_LEAK) of the leaks. The WLDIP model is a reliable, scalable, and cost-effective solution for improving water use efficiency in agricultural systems. Its implementation aligns with sustainable development goals related to clean water and responsible production, and it helps prevent adverse effects on natural ecosystems, particularly in semi-arid climatic zones. Future research could focus on automating this system for real-time leak monitoring.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"239 ","pages":"Article 110954"},"PeriodicalIF":8.9,"publicationDate":"2025-09-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144988003","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"DiffKNet-TL: Maize phenology monitoring with confidence-aware constrained diffusion model based on UAV platform","authors":"Guang Yang,&nbsp;Chang Liu,&nbsp;Gaoliang Li,&nbsp;Hong Chen,&nbsp;Keying Chen,&nbsp;Yakun Wang,&nbsp;Xiaotao Hu","doi":"10.1016/j.compag.2025.110936","DOIUrl":"10.1016/j.compag.2025.110936","url":null,"abstract":"<div><div>Agricultural phenological monitoring is key to understanding climate change impacts on ecosystems and optimizing crop management. UAV remote sensing provides high-resolution phenotypic data at the landscape scale, resolving plant-level phenological characteristics. However, weather and operational constraints hinder high-frequency UAV imaging on a daily scale. Previous studies have often focused on single phenotypic traits at single growth stages. Therefore, dynamic changes of multiple traits throughout the entire growth cycle were neglected, resulting in the identification of only partial phenological events. This study proposed a DiffKNet-TL (Diffusion-enhanced K-Net for tassels and leaves) model to identify maize phenology throughout the entire growth period, trained solely on tasseling stage images. A maize dataset covering leaves and tassels from seedling to harvest was also constructed. To capture dynamic phenotypic changes across scales and time, the K-Net architecture was also enhanced with a Transformer-based backbone and a dynamic loss reweighting strategy. For small size tassel targets, a confidence-aware discrete diffusion module was integrated, refining contours and reducing artifacts. Results showed that the implement of Swin-Transformer improved mIoU by 4.75%, and SSLossIoU added another 1.82%. Tassel refinement brought a further 2.55% IoU gain. Final IoUs for background, leaves, and tassels reached 84.98%, 75.85%, and 65.77%, respectively. DiffKNet-TL also generalized well across growth stages and remains robust under occlusion, straw residue, lighting disturbances, and uneven yellow leaf coloration. This study can provide the technical foundation and data support for large-scale, automated phenological monitoring of maize leaves and tassels, aiding in the quantitative dynamic tracking of maize phenology.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"239 ","pages":"Article 110936"},"PeriodicalIF":8.9,"publicationDate":"2025-09-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144988833","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Effect of agriculture-related dataset complexity on classical machine learning and deep learning classifiers performance","authors":"Gerardo Acevedo-Sánchez,&nbsp;Antonio Alarcón-Paredes,&nbsp;Cornelio Yáñez-Márquez","doi":"10.1016/j.compag.2025.110941","DOIUrl":"10.1016/j.compag.2025.110941","url":null,"abstract":"<div><div>This study evaluates how five indicators of dataset complexity affect the performance of 24 machine learning (ML) and deep learning (DL) classifiers across eight publicly available agriculture-related datasets. The indicators were cardinality (320–13,611 instances), dimensionality (7–35 features), class imbalance (Imbalance Ratio [IR] = 1–109.9), class number (2–40 classes), and feature types (numeric and ordinal). Performance measures, including sensitivity, specificity, balanced accuracy (BA), precision, F1-score, and Matthews Correlation Coefficient (MCC), were derived from confusion matrices generated via 10-fold cross-validation procedure. Macro and weighted-average were included as overall measures. Nonparametric tests (Friedman-Nemenyi; <em>p</em> &lt; 0.05 and Cliff’s [δ]) were performed for weighted-average sensitivity and BA. Across 192 analyses, ensembles (GBM, XGBoost, RF) and C5.0 significantly outperformed other classifiers on 5 out of 8 datasets, achieving values greater than 0.91. Artificial Neural Networks (ANN) showed ineffectiveness for tabular data (BA ≤ 0.50). Extreme imbalance (White Wine: IR = 109.9) affected the classifiers performance, mainly for distance-based and probabilistic (MCC ≤ 0.34), even the ensembles partially mitigated the bias (BA ≤ 0.65). High dimensionality (Date Fruits: 34 features) favored LDA and RF (BA ≥ 0.93). Conversely, large multiclass (Soybean Cultivars: 40 classes) demonstrated higher performance of IBk (BA = 0.87). Sixty paired comparisons confirmed significant differences (<em>p</em> &lt; 0.00001) and strong effects (δ = -0.57 to 0.18) between ensembles and underperforming classifiers, confirming that dimensionality, IR, and multiclass directly determine the performance. To the best of our knowledge, this is the first large-scale comparison of 24 ML/DL classifiers on eight agricultural datasets.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"239 ","pages":"Article 110941"},"PeriodicalIF":8.9,"publicationDate":"2025-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144933256","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Stability analysis of human hand grasping for the design of pneumatic muscle-driven end effector targeting citrus picking","authors":"Lina Wang ,&nbsp;Peigen Xu ,&nbsp;Jinbo Li ,&nbsp;Surkova Ekaterina ,&nbsp;Binrui Wang","doi":"10.1016/j.compag.2025.110942","DOIUrl":"10.1016/j.compag.2025.110942","url":null,"abstract":"<div><div>The integration of fruit-picking robots with anthropomorphic hand technology represents a significant advancement in agricultural automation and intelligence. The end effector is a key component for citrus harvesting robots. This paper proposes a new method to analyze and evaluate the stability of human hand grasping of citrus to guide the design of end effectors. A comprehensive assessment of 33 common human hand grasping postures was conducted, summarizing the 4 most commonly used human hand grasping types in citrus picking. Pressure data acquisition gloves were employed to obtain grasping force data from 16 regions, Principal Component Analysis algorithm was utilized to evaluate the functionality of each finger region, and wavelet transform algorithm was applied to assess the stability of the 4 grasping types. A pneumatic muscle-driven end effector for citrus picking was designed. The experimental results indicate that the grasping stability varies among different types. The stability ranking from high to low is as follows: Power palm-Thumb abduction type, Pinch type, Power pad-Thumb abduction type, and Power palm-Thumb adduction type. The stability of grasping direction from bottom to top is higher than that of side. The grasping performance test results show that the pneumatic muscle-driven end effector can stably grasp citrus fruits, with an average grasping time of 1.31 s and a grasping success rate of 100%. This study provides strong support for the design of end effectors for citrus picking robots.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"239 ","pages":"Article 110942"},"PeriodicalIF":8.9,"publicationDate":"2025-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144988820","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Robotic mowing of living crop cover as an alternative to plastic mulch: An approach for red fruit macro-tunnels","authors":"Sergio Bayano-Tejero ,&nbsp;Gregorio L. Blanco-Roldán ,&nbsp;Pedro Sánchez-Cachinero ,&nbsp;Rafael R. Sola-Guirado","doi":"10.1016/j.compag.2025.110924","DOIUrl":"10.1016/j.compag.2025.110924","url":null,"abstract":"<div><div>Plastic mulch is widely employed in macro-tunnel horticulture to suppress weeds, conserve soil moisture, and improve crop yield. There is a growing need for viable alternatives that mitigate its environmental impacts. One promising strategy is the introduction of vegetal cover crops, which function as living mulch. Nevertheless, the use of vegetative cover demands effective and frequent management to prevent competition with the main crop, presenting a key challenge in its widespread adoption. This study proposes and evaluates an autonomous robotic system designed for the mechanical management of vegetative cover as a sustainable alternative to plastic mulch in red fruit cultivation under macro-tunnel conditions. The robot integrates autonomous navigation and perception technologies built upon a ROS1 middleware framework. A custom-engineered mowing implement with reciprocating blades enables efficient and clean cutting, allowing plant residues to remain on the soil surface to maintain mulch functionality. The system has been validated through a combination of an artificial macro-tunnel and real-world field tests in macro-tunnel environments. Operational performance assessments indicated mowing coverage rates of 88.58 % and 84.55 % in artificial and real macro-tunnel, respectively. The average navigation error below 0.24 ± 0.02 m. The mowing evaluation reported a greater number and longer residues for the autonomous mowing robot (64.75 ± 12.84 and 131.17 ± 79.54 mm, respectively) compared to traditional systems (86.50 ± 9.37 and 101.05 ± 63.20 mm, respectively), which may contribute to slower decomposition and improve soil protection. The proposed robotic solution addresses key limitations in the manual maintenance of living mulch systems and offers a scalable, environmentally conscious approach to soil management.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"239 ","pages":"Article 110924"},"PeriodicalIF":8.9,"publicationDate":"2025-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144988822","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multi-feature image layers fusion for accurate detection of dairy cow mastitis using deep learning","authors":"Mengyuan Chu ,&nbsp;Yongsheng Si ,&nbsp;Ziruo Li ,&nbsp;Qian Li ,&nbsp;Gang Liu","doi":"10.1016/j.compag.2025.110937","DOIUrl":"10.1016/j.compag.2025.110937","url":null,"abstract":"<div><div>Mastitis in dairy cows is a costly disease that poses significant challenges to animal welfare and farm productivity. Traditional detection methods often rely on single feature and simple thresholding or machine learning algorithms, susceptible to environmental factors and individual cow specificity. To address these limitations, we proposed an automatic mastitis detection method based on the fusion of multiple feature image layers derived from thermal infrared imaging. First, thermal images of the udder region were captured to extract three distinct feature image layers, including temperature distribution, vascular structure, and udder size. The temperature layer captured the thermal variations of the udder and was generated by producing temperature contour plots from the raw images. The blood vessel layer highlighted vascular patterns derived from Laplacian filtering and skeletonization of the extracted udder region. The size layer provided measurements of udder dimensions, which were automatically detected using the CenterNet model and fitted with corresponding keypoints. These feature layers integrated both sides of the cow’s udder to create a dual-channel composite feature image. Finally, the DenseNet-201 deep learning model was employed to classify mastitis within a dataset of 7,000 thermal images. The proposed method achieved a classification accuracy of 91.88%, significantly outperforming the 78.60% accuracy obtained using only raw thermal images. Furthermore, ablation studies were conducted to assess the contributions of each feature layer to overall detection performance. The results demonstrate that multi-feature fusion achieved significantly superior performance to single features. Such an investigation provides a robust solution for early mastitis detection. It eliminates the requirements of manual feature extraction and temperature differential calculations, paving the way for automated animal health monitoring and unmanned farm management.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"239 ","pages":"Article 110937"},"PeriodicalIF":8.9,"publicationDate":"2025-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144933257","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}