Development of an automated fruit classification system by using computer vision and deep learning.

Beyond mean RTs in visual word recognition: Extensions of a remarkably stable three-way interaction amongst word frequency, stimulus degradation, and RT distributions.

Sequential viseme-driven visual speech recognition through dual-stream interactive neural architecture.

A lightweight four-channel multi-modal model to improve computational performance of automated fire detection

Computer vision offers significant potential for the continuous, stress-free, and cost-effective monitoring of animal behavior, yet its application in goat farming remains limited. In this study, a Multi-Object Detection (MOD) model was developed to classify goat behavior into four categories: eating, standing, drinking, and lying. Zenithal videos were recorded under both light and dark conditions on an experimental goat farm, resulting in 10,740 labelled annotations used to train, validate, and test 13 models leveraging Transformer- and CNN-based pretrained architectures. YOLO-based models (YOLOv8 and YOLOX) achieved the highest overall performances across both large and lightweight versions, demonstrating high detection capability and potential suitability for hardware-constrained scenarios. YOLOX-based MOD model is preferred for goat behavior detection due to its superior classification accuracy, fast inference speed, and fully open-source license, enabling flexible customization, deployment, and reproducibility. Other models, particularly DAB-DETR and H-DINO, underperformed, especially in detecting drinking behavior, which represents the most challenging class due to its visual similarity with standing, class imbalance, and fisheye distortion effects that affect the frame regions where drinkers are located. Mitigation strategies, including focal loss and distortion correction, improved detection accuracy for this class and reduced performance variability. The developed MOD model can be deployed for continuous group-level monitoring of goats, paving the way for scalable and efficient solutions for advanced behavioral analyses. Future works will focus on integrating tracking algorithms for animal-level insights, as well as on evaluating model generalizability across different farming conditions and goat breeds. • Computer vision applications for goat farming are still very limited in literature. • Various models for goat behavior detection were developed and tested. • YOLOX achieved a mAP @ 0.50 of 0.957 with strong performances across the classes. • Drinking detection is hard due to visual similarity, class imbalance, distortion. • Frame undistortion improves drinking detection performance and reduces variability.

To review recent advances in the use of artificial intelligence (AI) to address shortcomings in assessing and improving surgical performance/training by automating surgical skills assessment and feedback. We searched PubMed for studies published between 2015 and 2025 pertaining to AI for surgical training. Search terms included 'artificial intelligence or 'machine learning' or 'deep learning' and 'surgical feedback' or 'surgical training' or 'surgical skill'. Articles were identified with special attention given to those published in the last 5 years with a focus on AI for surgical skill assessment or feedback. Artificial intelligence has been used to successfully automate surgical skill assessment across a variety of surgical disciplines via approaches such as kinematics, sabermetrics, computer vision, and gesture analysis. Many of these studies have developed AI models capable of a binary classification of skill (novice vs expert), which demonstrate concordance when verified against ground truths from human raters. Based on these skills assessments, AI approaches may be further leveraged to generate automatic feedback, which has proven effective in improving surgeon performance metrics, particularly for underperformers. AI has also shown utility in categorising and analysing the content and impact of live surgical feedback, enabling more efficient analysis of how feedback can be best delivered to trainees. Artificial intelligence is a promising tool for augmenting surgical training and improving the objectivity and scalability of surgical skill assessment and feedback. To date, AI models are adept at detecting relatively large differences in surgical performance and providing rudimentary feedback. Further work is required to create models capable of doing more fine-tuned skill assessments and generating more detailed, constructive feedback.

Debates about whether development projects improve living conditions persist, partly because observational estimates can be biased by incomplete adjustment and because reliable outcome data are scarce at the neighborhood level. We address both issues in a continent-scale, sector-specific evaluation of Chinese and World Bank projects across 9899 neighborhoods in 36 African countries (2002-2013), representative of ∼ 88% of the population. First, we use a recent dataset that measures living conditions with a machine-learned wealth index derived from contemporaneous satellite imagery, yielding a consistent panel of 6.7 km square mosaics. Second, to strengthen identification, we proxy officials’ map-based placement criteria using pre-treatment daytime satellite images and fuse these with tabular covariates to estimate funder- and sector-specific ATEs via inverse-probability weighting. Incorporating imagery often shrinks effects relative to tabular-only models. On average, both donors raise wealth, with larger and more consistent gains for China; sector extremes in our sample include Trade and Tourism (330) for the World Bank (+12.29 IWI points), and Emergency Response (700) for China (+15.15). Assignment-mechanism analyses also show World Bank placement is often more predictable from imagery alone (as well as from tabular covariates). This suggests that Chinese project placements are more driven by non-visible, political, or event-driven factors than World Bank placements. To probe residual concerns about selection on observables, we also estimate within-neighborhood (unit) fixed-effects models at a spatial resolution about 67 times finer than prior fixed-effects analyses, leveraging the computer-vision-imputed IWI panels; these deliver smaller but, for Chinese projects, directionally consistent effects. Methodologically, we extend recent EO–ML causal inference frameworks by fusing pre-treatment satellite imagery with tabular covariates to estimate treatment propensities, and by systematically benchmarking image-augmented estimators against tabular-only and unit fixed-effects designs using new assignment-mechanism diagnostics. Empirically, we provide a continent-wide, sector-specific comparison of the neighborhood-level wealth effects of Chinese and World Bank projects across 9899 African neighborhoods. • Satellite imagery controls reveal that prior studies likely overstated aid benefits in Africa. • On average, Chinese development projects generate larger wealth gains than World Bank projects. • Earth Observation and machine learning improve causal estimates by proxying for planners’ maps. • China’s project placement is less predictable, suggesting unique political or event-driven logics. • Both donors raise wealth, but impacts vary widely by sector, with Emergency Response having the largest effect.

This article presents a dry fish image dataset framework to support data-driven research in computer vision, machine learning and Deep Learning. This dataset collected from a multiple market of dry fish in Dhaka, which shows that how different they look, how they are handled, and how they are presented in real world. Images were acquired using consumer-grade mobile cameras under natural lighting conditions to reflect practical deployment scenarios. The collection process purposefully incorporated variations in lighting, background clutter, camera angles, distances, and obstructions to augment data diversity. We got permission to take all the pictures and put them in a consistent naming and folder structure. The dataset has high-quality RGB images of twelve different types of dry fish, such as Bashpata, Chanda, Chapila, Chewa, Churi, Loitta, Shukna Feuwa, Shundori, Chingri, Kachki, Narkeli, and Puti Chepa. Each class includes dry fish species that are commonly traded and shows natural differences in size, texture, color, and drying patterns within the class. We looked at in the data and put them in groups by hand to make sure they were all in the same class by taking help of expert. We did some simple preprocessing, such as getting rid of duplicates and making sure that the formats were the same, all while keeping the data's original visual features.You can use this dry fish dataset again for things like classifying images, extracting features, analyzing data imbalance, benchmarking data augmentation, and visualizing explainable artificial intelligence. It could also help with research on how to recognize food with few resources, automate markets, digitize supply chains, and use mobile devices for inspections. The way the dataset is set up makes it easy to work with well-known deep learning frameworks. It can also be added to with more classes or metadata, making it useful for both academic research and practical development in smart food systems and fisheries informatics.

The AsianVehicle dataset introduces a comprehensive image collection of four traditional Bangladeshi vehicle types-Auto Rickshaw, Rickshaw, Rickshaw Van, and Leguna-captured to support research in computer vision and cultural informatics. These vehicles, integral to South and Southeast Asian transport systems, represent unique visual and structural characteristics seldom documented in existing global datasets. A total of 4000 RGB images were collected across various urban and semi-urban areas of Mirpur, Dhaka, using smartphone cameras under natural daylight conditions to preserve authentic colors, textures, and environmental diversity. The dataset encompasses variations in viewing angles, backgrounds, and illumination, reflecting real-world scenarios where such vehicles operate. All images are provided in properly processed form, enabling users to apply customized preprocessing, and labeling strategies according to their research needs. Beyond supporting machine learning tasks such as vehicle classification, segmentation, or detection, this dataset contributes to the digital preservation of traditional Asian transport designs that are gradually disappearing due to modernization. Its open accessibility facilitates comparative studies on model generalization, cross-domain adaptation, and low-resource visual recognition. By bridging cultural representation and artificial intelligence research, AsianVehicle offers a valuable foundation for both technical innovation and the preservation of regional identity within data-driven applications.

This study analyzes how the evaluation and classification of coffee are key processes to ensure the quality and commercial value of the product; However, traditional methods based on human inspection present limitations of subjectivity, time and consistency. In recent years, artificial intelligence has shown significant growth in applications aimed at automating these processes, especially through computer vision and machine learning. Given the rapid expansion and dispersion of these studies, the main motivation of this work is to systematically analyze and synthesize the advances in the use of artificial intelligence for the evaluation and classification of coffee. To this end, a systematic review was carried out following the PRISMA protocol, considering original studies published between 2021 and 2025 in scientific databases, a period selected to concentrate recent and consolidated developments. 30 articles were included, in which applications focused on grain quality control predominate, such as classification, defect detection and authenticity, using RGB images and NIR/UV-VIS spectra together with algorithms such as neural networks, SVM, Random Forest and XGBoost, with accuracies greater than 90%. Despite these results, challenges related to data availability, model generalization, and computational costs persist.

We introduce YOLIC Labeling, a tool designed to streamline the creation of high-quality datasets for cell wise object localization and classification. This tool integrates the Segment Anything Model (SAM) with customizable cell configurations (predefined regions of interest within images) to offer efficient and precise annotations. Key features include SAM-assisted labeling, manual polygon-based annotation, and semi-automatic labeling capabilities. By reducing manual labeling effort while maintaining accuracy, the tool supports the development of robust object localization and classification models, particularly those based on the You-Only-Look-at-Interested-Cells (YOLIC) methodology. YOLIC Labeling addresses the growing demand for efficient, versatile image annotation solutions in computer vision, with applications ranging from autonomous driving to smart industrial systems.

Lab-to-field integration in bridge monitoring: a hybrid structural health monitoring framework employing deep learning and unmanned aerial vehicle imagery

This review proposes candidate animal‑based indicators of good physical health and resilience in chickens (Gallus gallus domesticus) as a foundation for assessing positive animal welfare, complementing existing approaches to animal welfare assessment focused on use of iceberg indicators to detect severe health problems. We outline potential anatomical indicators involving the comb and wattles, eyes, beak, plumage, skin, footpads, claws, and overall body for rapid on-farm screening that could be automated for ease of application (e.g. using computer vision). We also identify health- and resilience-related anatomical and physiological indicators that could provide deeper, context‑dependent insights but require controlled testing conditions and/or laboratory analysis. For each indicator category, we summarize biological significance, influencing factors, and measurement methods under commercial and research settings. We classify candidate indicators according to their focus (health vs resilience) and response directionality on a scale from tolerable to optimal (whereby optimal values are highest for unidirectional measures such as plumage condition and intermediate for bidirectional measures such as claw length). We also rate potential ease of data collection (invasive, catching required, or remote sampling), on-farm applicability, and level of promise as a guide for indicator selection and prioritization for validation. Following validation and establishment of an appropriate scoring range from tolerable to optimal for each indicator depending on age, breed type, and reproductive status, we propose the use of continuous visual analogue scales or algorithms for scoring, followed by aggregation of indicator scores to obtain an overall rating of each bird's health and resilience. This narrative review thus provides a biologically grounded roadmap for developing proactive assessment tools that support thriving in poultry, as a foundation upon which affective and cognitive components of positive animal welfare can also be added.

A deep reinforcement learning algorithm for optimizing safety and efficiency of traffic signals using traffic conflict technique and artificial intelligence-based video analytics.

The changing landscape of concrete bridge infrastructure health monitoring and informed maintenance strategies: A decade of developments and trends

Deep learning-driven digital twin system for pedestrian tracking and evacuation load assessment in public spaces

Fish diseases remain a critical bottleneck for aquaculture, threatening fish welfare and causing significant economic losses. Current diagnostic approaches rely on expert visual inspection or laboratory analysis, which are time-consuming, labour-intensive, and often only confirm disease presence at advanced stages. By contrast, automated systems have the potential to detect early symptoms under real farming conditions quickly and at low cost once established. In this study, we tested two approaches (image-based and eDNA-based) that could be used as early warning systems for disease detection. We developed and validated a computer vision system for the early detection of red mark syndrome (RMS) in rainbow trout ( Oncorhynchus mykiss ). Using still images and underwater videos from controlled cohabitation experiments, we trained deep learning models (YOLOv11) to recognise early and late RMS lesions. On still image datasets, the system achieved mean Average Precision at 0.5 IoU (mAP50, IoU - Intersection over Union) values of 0.87–0.92, while on underwater video frames, it reached 0.92, further improving to 0.96 after adding heuristic rules. Comparison with expert evaluations showed similar scoring of disease characteristics, demonstrating that the automated approach can replicate expert-level assessments. Parallel monitoring of environmental DNA for the suspected causative bacterial agent of RMS ( Midichloria -like organism) revealed only low-level signals, mostly in later stages, indicating that eDNA is not a reliable early indicator of RMS under flow-through aquaculture conditions. Our results demonstrate that automated computer vision can provide accurate, real-time, and remote detection of early RMS symptoms, offering a scalable early-warning system to support proactive fish health management and reduce disease impacts in aquaculture. • Automated computer vision detected early RMS lesions in rainbow trout with high accuracy (mAP50 up to 0.92 on images and 0.96 on video). • Automated lesion counts and colour trends closely matched expert assessments of disease progression. • eDNA monitoring detected the causative Midichloria-like organism only at low levels and mainly during late disease stages. • The image-based RMS signs detection can be adapted to any disease with visible signs • Image-based monitoring provides a reliable, real-time early warning system applicable under production aquaculture conditions.

Automated measurement of aortic parameters using deep learning and computer vision

Computational models reveal intuitive physics and statistical cues separately contribute to the visual perception of liquids.

Assessing the nest building activity of ground-nesting bees by surveilling tumuli changes using camouflaged instance segmentation