Computers and Electronics in Agriculture最新文献

{"title":"Harnessing deep learning for plant species classification: A comprehensive review","authors":"Aisha Zulfiqar,&nbsp;Ebroul Izquierdo,&nbsp;Krishna Chandramouli","doi":"10.1016/j.compag.2025.110663","DOIUrl":"10.1016/j.compag.2025.110663","url":null,"abstract":"<div><div>Plants are fundamental to Earth’s ecosystems and human life, playing an indispensable role in sustaining and improving the global ecological balance. Understanding plant biodiversity, including species distribution and population dynamics, is essential for ecological and environmental protection. Automated plant species classification, driven by advanced machine learning and computer vision technologies, is a key step towards biodiversity conservation. This survey presents a comprehensive review of the past decade’s research on automated plant species classification, focusing on datasets and identification methods. It highlights the progression from single-organ, single label species classification to multi-organ, multi-label approaches for more comprehensive biodiversity monitoring. The study examines deep learning approaches, emphasizing moving from supervised learning to semi-supervised, self-supervised, and few-shot learning paradigms. It also highlights the progression from Convolutional Neural Networks to Vision Transformers and the performance evaluation on six plant species datasets. A notable contribution of this study is the evaluation of state-of-the-art few-shot learning methods on six plant datasets. By synthesizing trends, key findings, and critical analyses, this paper offers valuable insights into past advancements and identifies future research directions and challenges, paving the way for enhanced automated plant species classification and biodiversity assessment.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"237 ","pages":"Article 110663"},"PeriodicalIF":7.7,"publicationDate":"2025-07-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144596484","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":"TDR-Transformer: A transformer neural network model to determine soil relative permittivity variations along a time domain reflectometry sensor waveguide","authors":"Zhuangji Wang ,&nbsp;Dennis Timlin ,&nbsp;Xiaofei Gong ,&nbsp;Yuki Kojima ,&nbsp;Shan Hua ,&nbsp;David Fleisher ,&nbsp;Wenguang Sun ,&nbsp;Sahila Beegum ,&nbsp;Vangimalla R. Reddy ,&nbsp;Katherine Tully ,&nbsp;Robert Horton","doi":"10.1016/j.compag.2025.110730","DOIUrl":"10.1016/j.compag.2025.110730","url":null,"abstract":"<div><div>Interpreting soil relative permittivity (<span><math><mrow><msub><mi>ε</mi><mi>r</mi></msub></mrow></math></span>) variations along a time domain reflectometry (TDR) waveguide provides an opportunity to determine soil water content at multiple depths using a vertically installed TDR sensor. Compared to placing sensors at different depths, vertical sensor installation reduces measurement efforts and enhances data-use-efficiency. Revealing <span><math><mrow><msub><mi>ε</mi><mi>r</mi></msub></mrow></math></span> variations includes two aspects: identifying <span><math><mrow><msub><mi>ε</mi><mi>r</mi></msub></mrow></math></span> change positions and determining <span><math><mrow><msub><mi>ε</mi><mi>r</mi></msub></mrow></math></span> values. Traditional inverse analyses are not widely applied due to their high computational demands. Machine learning-based methods, e.g., TDR-CNN, provide a forward computational workflow to track <span><math><mrow><msub><mi>ε</mi><mi>r</mi></msub></mrow></math></span> change positions and reduce computational load, but errors in <span><math><mrow><msub><mi>ε</mi><mi>r</mi></msub></mrow></math></span> values are relatively large. In this study, TDR-Transformer is developed as a new waveform interpretation model to improve <span><math><mrow><msub><mi>ε</mi><mi>r</mi></msub></mrow></math></span> estimation accuracy. Modified from the standard transformer architecture, an encoder with convolutional neural layers is used to extract waveform geometric features, and a decoder generates a sequence of <span><math><mrow><msub><mi>ε</mi><mi>r</mi></msub></mrow></math></span> values to represent <span><math><mrow><msub><mi>ε</mi><mi>r</mi></msub></mrow></math></span> variations. Attention is a mechanism that can dynamically extract and process the relevant information within the data, which processes and integrates the waveform geometric information in the encoder, ensures the causality (time-order) of the waveform data in the decoder, and transfers information from the encoder to the decoder. TDR-Transformer was trained and tested using simulated waveforms where <span><math><mrow><msub><mi>ε</mi><mi>r</mi></msub></mrow></math></span> changes along the waveguides, but soil electrical conductivity (EC) was assumed to be small and stable. The RMSE for <span><math><mrow><msub><mi>ε</mi><mi>r</mi></msub></mrow></math></span> values was within 0.5–1.6 % and the RMSE of <span><math><mrow><msub><mi>ε</mi><mi>r</mi></msub></mrow></math></span> change positions was within 5–8 %. A soil infiltration experiment and a precipitation-evaporation experiment illustrated applications of TDR-Transformer to observed waveforms. Consequently, TDR-Transformer is a promising artificial intelligence model to interpret TDR waveforms in soils with nonuniform <span><math><mrow><msub><mi>ε</mi><mi>r</mi></msub></mrow></math></span>, and fine-tuning TDR-Transformer is recommended for specific commercial TDR sensor designs.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"237 ","pages":"Article 110730"},"PeriodicalIF":7.7,"publicationDate":"2025-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144581166","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 vision-based pruning algorithm for cherry tree structure elements segmentation and exact pruning points determination","authors":"Axios Kefalas,&nbsp;Theofanis Kalampokas,&nbsp;Eleni Vrochidou,&nbsp;George A. Papakostas","doi":"10.1016/j.compag.2025.110735","DOIUrl":"10.1016/j.compag.2025.110735","url":null,"abstract":"<div><div>The lack of adequately skilled seasonal workforce for agricultural precision tasks such as pruning boosted the development towards automation. Robotic tree pruning, however, requires high precision in determining pruning points and needs to be based on specific pruning practices to be effective. This work presents the first machine vision-based complete algorithm following strict and precise pruning rules for dormant cherry trees of the Central Leader training system, aiming to guide automated cherry pruning. First, multi-class semantic segmentation is performed by testing U-Net with three different feature extraction backbones, to detect the best performing combination. Then, geometrical calculations based on specific pruning strategies are employed to locate the exact cutting point on the detected trunks, branches and shoots. Segmentation results reported an IoU of 98.5 % for three classes (trunk, branches, shoots) by using U-Net with VGG16. We also validated the performance of cutting points determination method, achieving an average accuracy rate of 93.33 %, reporting 88.75 % precision for cutting points on branches, 91.25 % on shoots, and 100 % on trunks, across eight trees. The proposed methodology is the first in the bibliography to propose a vision-based precision pruning algorithm, based on strict pruning rules, that moreover determines the exact pruning points for the Central Leader Training System for dormant cherry trees. Moreover, the adopted pruning strategy can be used for the annual formulation of tree shape, aiming to cover all types of selective pruning tasks, while it can be easily adapted to fit the pruning practices of other tree types by modifying the pruning rules.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"237 ","pages":"Article 110735"},"PeriodicalIF":7.7,"publicationDate":"2025-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144581164","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":"Early yield estimation in wheat using open access global datasets and artificial intelligence","authors":"Muhammed Cem Akcapınar ,&nbsp;Halit Apaydin","doi":"10.1016/j.compag.2025.110486","DOIUrl":"10.1016/j.compag.2025.110486","url":null,"abstract":"<div><div>The determination of the production amount of wheat, a staple foodstuff widely cultivated across the globe, is of significant importance prior to the harvest. This is due to the fact that the price of wheat is dependent upon the supply and demand balance, and therefore, accurate forecasting of production is essential for effective food production planning. A great deal of research has been conducted for this purpose. However, in addition to the fact that agricultural production is highly dependent on environmental factors, the diversity of data affecting the amount of production, the change in the growing season and the constraints on its continuity can present a significant challenge. In instances where data cannot be sourced from local data sources or there is a lack of homogeneity, researchers may turn to alternative data sources. In this context, digital products, particularly those developed using satellite technology, have begun to be employed in agricultural contexts, such as yield estimation studies, and have assumed an important role in addressing this need. This study aims to develop artificial intelligence-based wheat yield prediction models for Ankara province in Türkiye by utilizing a range of global datasets that are accessible to international researchers. The models were populated with a number of input parameters, including the Spatial Production Allocation Model (SPAM) wheat-masked Normalized Difference Vegetation Index (NDVI), the Famine Early Warning Systems Network Land Data Assimilation System (FLDAS) evapotranspiration, the Moderate Resolution Imaging Spectroradiometer (MODIS) Land Surface Temperature (LST) and the European Centre for Medium-Term Weather Forecasting (ECMWF) AgERA5 precipitation data. The study was conducted using four artificial intelligence-based algorithms (Support Vector Regression (SVR), Random Forest (RF), Artificial Neural Networks (ANN) and Long Short-Term Memory (LSTM)) in the Google Colaboratory environment. A total of 180 models were developed using these algorithms, comprising 15 different scenarios and three distinct approaches to optimizing model performance, facilitating early prediction and reducing the number of variables employed. The most successful model was identified as the one developed with the LSTM algorithm, which achieved an R² value of approximately 0.96. The models developed with the ANN, RF, and SVR algorithms exhibited comparatively lower success, with R² values of 0.86, 0.84, and 0.76, respectively. Furthermore, the model developed with LSTM offers forecasting capabilities at least three months prior to the harvest period. The study concluded that successful yield estimation can be achieved through the utilization of open access international datasets, particularly in scenarios where agricultural data are scarce.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"237 ","pages":"Article 110486"},"PeriodicalIF":7.7,"publicationDate":"2025-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144581165","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":"ROI-aware multiscale cross-attention vision transformer for pest image identification","authors":"Ga-Eun Kim ,&nbsp;Chang-Hwan Son ,&nbsp;Soojeong Lee","doi":"10.1016/j.compag.2025.110732","DOIUrl":"10.1016/j.compag.2025.110732","url":null,"abstract":"<div><div>The pests captured with imaging devices may be relatively small in size compared to the entire images, and complex backgrounds have colors and textures similar to those of the pests, which hinders accurate feature extraction and makes pest identification challenging. The key to pest identification is to create a model capable of detecting regions of interest (ROIs) and transforming them into better ones for attention and discriminative learning. To address these problems, we will study how to generate and update the ROIs via multiscale cross-attention fusion as well as how to be highly robust to complex backgrounds and scale problems. Therefore, we propose a novel ROI-aware multiscale cross-attention vision transformer (ROI-ViT). The proposed ROI-ViT is designed using dual branches, called Pest and ROI branches, which take different types of maps as input: Pest images and ROI maps. To render such ROI maps, ROI generators are built using soft segmentation and a class activation map and then integrated into the ROI-ViT backbone. Additionally, in the dual branch, complementary feature fusion and multiscale hierarchies are implemented via a novel multiscale cross-attention fusion. The class token from the Pest branch is exchanged with the patch tokens from the ROI branch, and vice versa. The experimental results show that the proposed ROI-ViT achieves 81.81%, 99.64%, and 84.66% for IP102, D0, and SauTeg pest datasets, respectively, outperforming state-of-the-art (SOTA) models, such as MViT, PVT, DeiT, Swin-ViT, and EfficientNet. More importantly, for the new challenging dataset IP102(CBSS) that contains only pest images with complex backgrounds and small sizes, the proposed model can maintain high recognition accuracy, whereas that of other SOTA models decrease sharply, demonstrating that our model is more robust to complex background and scale problems.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"237 ","pages":"Article 110732"},"PeriodicalIF":7.7,"publicationDate":"2025-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144572512","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":"Real-time pose estimation of oriental melon fruit-pedicel pairs using weakly localized fruit regions via class activation map","authors":"Seung-Woo Kang ,&nbsp;Kyoung-Chul Kim ,&nbsp;Yong-Joo Kim ,&nbsp;Dae-Hyun Lee","doi":"10.1016/j.compag.2025.110734","DOIUrl":"10.1016/j.compag.2025.110734","url":null,"abstract":"<div><div>Fruit pose is an important trait for robotic harvesting that provides not only control information for the robot arm but also indicates whether the fruit can be harvested. A pair of fruit and pedicel has been frequently used as the target object for estimating the fruit pose, and the pose estimation of multiple fruit-pedicel pairs within an image is also being focused on in current research to determine the targets to be harvested efficiently. The multi-fruit pose estimation is performed based on accurate object detection that requires a large amount of training data for feature representation of the object, especially pedicel, in complex background; however, there are spatio-temporal constraints in collecting various fruit-pedicel images, and precise labelling is a laborious task. Easier feature representation for localization can reduce model complexity and data-scale dependency. In this study, multi-fruit pose estimation of oriental melon was proposed, which includes a weakly fruit localization that reduces labelling effort while providing the fruit region containing the essential information for estimation of fruit-pedicel pose. The framework consists of a weak supervision-based localization and pose estimation models connected in sequence. The fruit regions are simultaneously and approximately localized, and each of these regions is then fed sequentially to the pose estimation. The proposed method has the advantage of being able to use a smaller model than existing models, which allows for achieving performance even with a small data scale and also improving real-time performance. The results showed that, compared to YOLO-pose models, our method achieved up to 0.23 higher PDK scores and exhibited an inference speed improvement of up to 8 fps. Our method minimized the effort required for localization and improved speed while demonstrating performance comparable to existing studies.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"237 ","pages":"Article 110734"},"PeriodicalIF":7.7,"publicationDate":"2025-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144581280","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":"Deep metric learning for soil organic matter prediction: A novel similarity-based approach using smartphone-captured images","authors":"Mojtaba Naeimi ,&nbsp;Vishvam Porwal ,&nbsp;Stacey Scott ,&nbsp;Maja Krzic ,&nbsp;Prasad Daggupati ,&nbsp;Hiteshkumar Vasava ,&nbsp;Daniel Saurette ,&nbsp;Ayan Biswas ,&nbsp;Abhinandan Roul ,&nbsp;Asim Biswas","doi":"10.1016/j.compag.2025.110728","DOIUrl":"10.1016/j.compag.2025.110728","url":null,"abstract":"<div><div>The accurate assessment of soil organic matter (SOM) is crucial for sustainable agriculture, yet traditional methods remain time-consuming and costly. While smartphone-based digital imaging offers a promising alternative, current approaches face limitations in prediction reliability and generalization capability. This study introduces a novel similarity-based deep learning framework for SOM prediction using smartphone-captured soil images, fundamentally shifting from traditional regression-based methods to a metric learning paradigm. We developed an enhanced image acquisition system and implemented a Triplet Loss network architecture that learns to embed soil images in a semantic space where similarity relationships correlate with SOM content. The system incorporates adaptive image quality assessment and enhancement using the Blind/Referenceless Image Spatial Quality Evaluator and super-resolution techniques. Experimental validation using 500 soil samples from Southern Ontario demonstrated superior performance of our similarity-based approach (validation RMSE = 0.17) compared to traditional regression methods (validation RMSE = 0.51 for Random Forest). The model maintained consistent performance across different soil textures (RMSE variation &lt; 0.05 between texture classes) and environmental conditions (temperature 20–30 °C, humidity 45–75 % RH). The complete analysis pipeline makes the system practical for field applications. Our approach addresses critical challenges in digital soil analysis by providing rapid, reliable, and accessible SOM assessment, contributing to improved soil monitoring and management practices in precision agriculture. These findings demonstrate the potential of similarity-based learning for advancing digital soil sensing technologies and supporting sustainable agricultural practices.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"237 ","pages":"Article 110728"},"PeriodicalIF":7.7,"publicationDate":"2025-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144572350","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":"TinySeg: A deep learning model for small target segmentation of grape pedicels with multi-attention and multi-scale feature fusion","authors":"Yue Wu ,&nbsp;Xinjie Yu ,&nbsp;Derong Zhang ,&nbsp;Yong Yang ,&nbsp;Yi Qiu ,&nbsp;Limin Pang ,&nbsp;Hangjian Wang","doi":"10.1016/j.compag.2025.110726","DOIUrl":"10.1016/j.compag.2025.110726","url":null,"abstract":"<div><div>Accurate pedicel identification is critical for advancing the automation of table grape harvesting processes. The pedicel is a typical small-target recognition scenario that poses significant challenges. This study proposes an instance segmentation network named TinySeg, aiming at enhancing the recognition accuracy of small pedicel targets. It comprises a MobileFormer feature extraction module, a TriFPN feature fusion module, and multiple parallel prediction heads. The MobileFormer incorporates spatial, channel, and self-attention mechanisms to enhance focus on small targets and integrates an SPD-Conv module to expand the receptive field while preserving fine-grained details. The TriFPN network is designed to simultaneously capture detailed information from lower-level feature maps and semantic information from higher-level feature maps, providing richer contextual information for the detection of small targets. Furthermore, an adaptive loss function is introduced to alleviate the class imbalance between large and small target samples, which significantly boosts the model’s training efficiency. Experiments demonstrate that TinySeg outperforms mainstream instance segmentation networks in recognizing small pedicel targets, achieving precision, recall, mAP, and mIoU of 87.1%, 74.2%, 80.7%, and 76.6%, respectively. Compared to the best-performing SOLOv2, TinySeg improves mAP and mIoU by 6.1% and 5.5%, respectively. In terms of model complexity, TinySeg boasts only one-seventh the number of parameters of SOLOv2 and achieves nearly double its inference speed, making it a highly promising solution for practical applications in table grape harvesting automation.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"237 ","pages":"Article 110726"},"PeriodicalIF":7.7,"publicationDate":"2025-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144572349","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":"Predicting the 3D bone structure of pork shoulders using X-ray imaging and statistical shape modeling","authors":"Michiel Pieters ,&nbsp;Pieter Verboven ,&nbsp;Bart M. Nicolaï","doi":"10.1016/j.compag.2025.110666","DOIUrl":"10.1016/j.compag.2025.110666","url":null,"abstract":"<div><div>In the meat industry, deboning and cutting are essential processing steps often performed by humans. To automate these processes it is essential to understand the complexity and variability of their three-dimensional (3D) shape as well as the relationship between their outer shape and inside bone structure. In this paper, we introduce a 3D statistical shape model (SSM) that describes the outer surface of a pork shoulder and its corresponding inner bone structure. X-ray computed tomography (CT) scans were acquired from 45 right-hand side and 45 left-hand side pork shoulders. The CT scans were segmented to obtain 3D models of the external shape and internal bone structure. Surface meshes were then created and used for establishing SSMs of the outer surface, the bone structure and the combined surfaces of the left and right pork shoulders based on principal component analysis. The first five of a total of 40 principal components were able to describe 63.6 % of the variability in the entire dataset. The mean absolute error (MAE) of the proposed fitting method in this paper was 5.15 mm for the test set. Besides being compact, the models could also generate realistic 3D shapes of pork shoulders that were not present in the dataset. These shapes can be used in the development of automated cutting and deboning procedures, and, thus, lead to improved precision in cutting, reduced waste, and further enhancements of automation within the meat industry.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"237 ","pages":"Article 110666"},"PeriodicalIF":7.7,"publicationDate":"2025-07-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144570757","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":"Advancing animal farming with deep learning: A systematic review","authors":"Zahid Ur Rahman,&nbsp;Mohd Shahrimie Mohd Asaari,&nbsp;Haidi Ibrahim","doi":"10.1016/j.compag.2025.110674","DOIUrl":"10.1016/j.compag.2025.110674","url":null,"abstract":"<div><div>Deep learning has revolutionized animal farming by enabling automated health monitoring, behavior analysis, and livestock management. This review examines the application of key deep learning architectures, including convolutional neural networks (CNNs), You Only Look Once (YOLO), memory-based neural networks (MBNNs), and generative adversarial networks (GANs), in various aspects of animal farming. These models have demonstrated success in tasks such as real-time livestock detection, disease prediction, activity monitoring, and animal identification. However, challenges such as occlusion, data scarcity, small training datasets, environmental variability, and imbalanced data remain significant barriers to model reliability and scalability. By analyzing one hundred seventeen articles following PRISMA guidelines, this review highlights recent advancements, identifies research gaps, and discusses solutions such as image augmentation, synthetic data generation, and domain adaptation. The findings emphasize the potential of deep learning to enhance precision farming while addressing critical challenges. Finally, future research directions are proposed to improve model generalization, integration with IoT-based monitoring systems, and real-time decision-making for sustainable and intelligent livestock management.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"237 ","pages":"Article 110674"},"PeriodicalIF":7.7,"publicationDate":"2025-07-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144570758","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}