Computer vision models for precision poultry farming: A narrative review of behavioral and welfare monitoring studies.

Spatial-temporal risk field-based coupled dynamic-static driving risk assessment and trajectory planning in weaving segments.

Acquiring and processing full-motion details in machine vision typically consumes a substantial amount of energy. In contrast, a hierarchical processing architecture, combining a low-power standby front end with an on-demand activated back end, provides an optimized energy-performance tradeoff. To achieve this, the complete acquisition and decoupling of static (brightness) and dynamic (amplitude and polarity) output at the sensory level are essential for activating on-demand vision function. Here, we report a differential image sensor (DIS) that leverages differential photodiodes with decoupled differential and tunneling modes. These modes can be read out via conventional ROICs, paving the way for the up-scaled integration (e.g., 640 × 512). With on-demand activated dynamic and static modes, the DIS implements a hierarchical motion-processing pipeline-from sparse motion detection to optical flow and depth analysis. This work provides a power-efficient and scalable strategy for advancing vision-based AIoT applications.

Gorgon Loop is an interactive art installation that examines how AI-driven judgement systems operate in public space and how they simulate and amplify collective discourse and social gaze. The name Gorgon Loop draws on Greek mythology: like Medusa's paralysing gaze, machine vision implies power, while Medusa — shaped by projection rather than born monstrous — mirrors how AI is granted false agency. Comprising five rotatable “intelligent mirrors,” the system activates when viewers enter the installation area. Using machine vision, it translates appearance and posture into structured parameters that are both displayed on the mirror surfaces and used as input for generative language agents. These agents produce sequential, persona-constrained text commentary with evaluative overtones, unfolding across the mirrors in a format resembling a group chat. By exposing intermediate visual features, extracted descriptors, and the resulting language output as a continuous public performance, Gorgon Loop renders classification, inference, and bias perceptible as lived experience rather than hidden technical process.

The rapid advancement of artificial intelligence has propelled the development of β-Ga2O3 photo-synapses for solar-blind ultraviolet neuromorphic machine vision systems. However, existing β-Ga2O3 photo-synapses not only exhibit reduced stability but also display high weight update nonlinearity. Herein, we propose a novel strategy to construct β-Ga2O3 photo-synapses with low weight update nonlinearity based on self-trapped holes, aiming to achieve multi-level in-sensor computing tasks. Theoretical and experimental investigations revealed that the interaction between the larger effective mass of holes and local lattice distortions in β-Ga2O3 promoted the formation of self-trapped holes, which significantly reduced hole mobility and enhanced the persistent photocurrent effect. The fabricated β-Ga2O3 photo-synapses exhibited excellent short-term plasticity, which could be transited to long-term plasticity by adjusting the characteristics of 252 nm ultraviolet light. Moreover, the devices achieved a low weight update nonlinearity of 0.42, outperforming most previously reported photo-synapses. Finally, β-Ga2O3 photo-synapses were integrated into neuromorphic machine vision systems, enabling tasks ranging from low-level image classification to high-level motion recognition, achieving recognition accuracies of 99.48% and 92.70% on the MNIST and Fashion-MNIST datasets. It also maintained 100% target tracking accuracy under 60% Gaussian noise interference and reached a recognition accuracy of 94.94% for 10 motions in UTD-MHAD dataset. These results highlight great potential of β-Ga2O3 photo-synapses based on self-trapped holes engineering in the era of artificial intelligence.

ABSTRACT Purpose This study aimed to report the longitudinal Ora Visual Navigation Course (Ora VNC™) mobility test results in two individuals, with CEP290 LCA following intravitreal injections of an antisense oligonucleotide, sepofarsen. Methods Two individuals were enrolled in the Illuminate Phase 3 trial and subsequently in the Post-Trial Access (PTA) program, undergoing Ora VNC™ across 12 visits. Results Subject 1 was initially randomized to the control (sham) group and switched to treatment from control group from the month 12 visit onwards. This individual passed only a few courses but demonstrated signs of enhanced spatial perception (defined by, among others, the ability to perceive objects’ shapes and sizes). FST (full-field stimulus threshold) supported the light sensitivity gain in this subject. In contrast, subject 2, who received treatment since baseline visit, successfully completed mobility tests in all visits with progressive improvement, reaching lower luminance and more challenging contrast settings over time. According to the technicians’ report, both subjects demonstrated an increase in perceived obstacle recognition during iterations, which may reflect improvement of mobility-based visual function outcomes. Conclusion The Ora VNC™ mobility test captured mobility-based visual function gains, including subtle spatial improvements, but may lack sensitivity in individuals with extremely low vision, potentially overlooking small but patient-relevant changes

Sorghum is an important multi-purpose crop used for food, feed, brewing, and bioenergy. However, mechanised harvesting is hindered by its tall stature, complex panicle morphology, small and fragile grains, high-moisture stems and leaves, and susceptibility to lodging at maturity. This review provides a PRISMA-guided systematic literature search and narrative synthesis of mechanised sorghum harvesting from crop adaptability to intelligent equipment. The main literature search covered publications from January 1990 to May 2026 and included Web of Science Core Collection, Scopus, ScienceDirect, SpringerLink, Google Scholar, and other relevant sources. A total of 1928 records were identified, and 190 studies were finally included in the qualitative synthesis after duplicate removal, title-and-abstract screening, and full-text assessment. The review analyses how plant morphology, panicle exsertion, physical and mechanical properties, maturity stage, moisture content, varietal differences, and lodging affect harvesting suitability. The applicable conditions for segmented harvesting, panicle harvesting, direct grain harvesting, and multi-purpose coordinated harvesting are compared under different crop, regional, and machinery conditions. Key technological advances are synthesised in relation to header feeding, threshing and cleaning, straw management, and operating-parameter optimisation. Recent developments in machine vision, condition monitoring, adaptive control, DEM and CFD-DEM simulation, and digital twins are also assessed. The analysis shows that high-quality mechanised sorghum harvesting requires the coordinated optimisation of crop traits, harvesting methods, machine structures, operating parameters, and digital sensing–control feedback. The main contribution of this review is to establish an integrated crop–machine–operation framework for identifying technical constraints, comparing harvesting routes, and guiding the development of specialised, low-loss, low-breakage, and intelligent sorghum harvesting systems.

Driven by the rapid progress of artificial intelligence and robotics, neuromorphic vision systems are gaining significant attention for enabling efficient visual information processing in complex and dynamic environments. In particular, optoelectronic devices that emulate the functionality of the biological retina are essential for achieving efficient neuromorphic visual processing. Here, we report an optoelectronic synaptic memtransistor (OSMT)-based neuromorphic vision system for image processing applications. By integrating photoresponsive indium-gallium-zinc-oxide (IGZO) as the channel material with a hafnium oxide (HfO2) contact-engineered architecture, the OSMT exhibits optically and electrically tunable resistive states, enabling stable and controllable synaptic weight modulation for artificial neural network (ANN) implementation. Benefiting from reliable optoelectronic synaptic characteristics, ANN simulations achieve a handwritten digit recognition accuracy of 92.17%. Furthermore, a 6×6 OSMT array demonstrates neuromorphic image processing capabilities, including contrast enhancement. These results highlight the potential of OSMTs as key building blocks for intelligent machine vision systems, offering new opportunities for advanced robotic platforms and human-machine interfaces.

Reasoning is a hallmark of human intelligence, enabling adaptive decision-making in complex unfamiliar scenarios. In contrast, machine intelligence remains bound to training data, unable to dynamically refine solutions at inference. While recent advances have explored machine reasoning - trading inference-time compute for improved performance - they focus on verbal domains such as mathematical problem-solving where explicit rules govern step-by-step solution generation. Many tasks lack sufficient labelled data and require alternative performance improvement mechanisms, such as inference-time compute. Here we present a paradigm for machine reasoning in vision, enabling performance improvements with increasing thinking time (inference-time compute), even with limited labelled data. Our approach is inspired by dual-process theories of human cognition, integrating a fast-thinking System I module for generating and verifying solutions in familiar tasks, with a slow-thinking System II module that iteratively refines predictions using self-play reinforcement learning, even when task-specific data is limited. This paradigm involves proposing, competing over, and refining solutions until convergence. We demonstrate that extended inference-time compute yields superior performance compared to large-scale supervised learning, foundation models, and human experts in vision tasks. These include computer-vision benchmarks and cancer localisation across five organs, highlighting the potential of inference-time compute for data-scarce problems.

Color difference detection remains a critical challenge in textile manufacturing, where traditional visual inspection and offline measurement methods suffer from subjectivity, low efficiency, and delayed feedback. This study emphasizes engineering integration for online industrial fabric inspection rather than proposing a single new color-difference algorithm. The proposed system integrates a custom-designed optical acquisition platform with a lightweight color analysis pipeline, including bilateral filtering for noise suppression, K-means clustering for representative color extraction, RGB-to-CIELab color space conversion, and perceptually weighted [Formula: see text] computation. The system was deployed on an actual textile production line and evaluated using ten fabric rolls with different colors and materials. Experimental results show roll-level agreement with manual inspection in the tested samples and indicate the feasibility of continuous monitoring of chromatic variations along the fabric length. The proposed system provides a practical engineering solution for automated textile color quality control and may support production-line decision making while reducing dependence on subjective visual inspection in industrial environments.

Fruit harvesting systems are undergoing a paradigm shift toward sustainable and energy-efficient mechanized platforms driven by robotics, artificial intelligence, and advanced sensing technologies. This review synthesizes recent engineering developments in fruit harvesting, focusing on system architecture, fruit detachment mechanics, and mechanized harvesting strategies. It examines harvesting classifications, mechanical principles governing detachment, and pre-harvest factors affecting performance, along with principal mechanisms including shaking, cutting, and alternative detachment techniques. Post-detachment handling and fruit recovery processes are also analyzed, together with economic and sustainability-related trade-offs between manual and mechanized harvesting systems. Recent progress in robotic harvesting systems, machine vision, and multi-sensor fusion is evaluated within the framework of smart orchard engineering, with increasing emphasis on energy-efficient design, resource optimization, reduced postharvest losses, and environmental sustainability as key performance drivers. Despite these advancements, current technologies remain constrained by fruit damage susceptibility, biological variability, limited cross-crop adaptability, and high implementation costs, limiting large-scale adoption in commercial orchards. The novelty of this review lies in establishing a unified engineering framework that links mechanical detachment principles with robotic systems and intelligent sensing technologies under an energy-efficient sustainability perspective, enabling a system-level understanding of harvesting performance and supporting the development of next-generation adaptive and sustainable fruit harvesting systems.

Machine vision systems have been increasingly adopted for defect inspection and classification in manufacturing. Beyond identifying pre-defined defect categories observed during training, these systems are desired to also detect new defect types that are Out-of-Distribution (OoD), which differs significantly from In-Distribution (InD) training data. Existing DNN-based OoD detection methods rely on predictive uncertainty or regularization on limited OoD samples, but they often overfit specific OoD samples and fail to generalize. To overcome these limitations, this article proposes a novel generative adversarial approach that uses limited real OoD samples for supervised OoD sample generation with exploration of OoD space. Our method tackles two major challenges of existing works: first, it provides a supervised OoD generation scheme based on real OoD samples, unlike existing unsupervised methods that generate virtual outliers based on InD data; second, it simultaneously augments OoD samples and explores unseen OoD space, reducing the issues of insufficient real OoD samples and overfitting. The article provides rigorous theoretical results to demonstrate the efficacy of the proposed approach. Comparison studies with state-of-the-art techniques demonstrate superior generalizability to unseen OoD data. Furthermore, we showcase its practical effectiveness in detecting new surface defects using a real-world 3D point cloud dataset of manufacturing defects.

Optoelectronic synaptic devices enable in-sensor processing of enhanced edge detection and contrast resolution in complex visual scenes due to their excellent capability to emulate the functions of visual neurons, such as light perception and image processing, while lateral inhibition synaptic plasticity refines spatial selectivity and extends the dynamic range by suppressing redundant signals and amplifying subtle variations in input intensity. The incorporation of lateral inhibition into a single optoelectronic synaptic device will offer a cost-effective and energy-efficient route for directing a robotic arm to perform responding motions and developing highly efficient machine vision systems. Herein, we demonstrate an optoelectronic artificial synapse established on a novel heterostructure consisting of metal oxide In2O3, polycrystalline Cs2AgBiBr6perovskite, and indium-gallium-zinc oxide thin film, which enhances the optoelectronic response and corresponding synaptic plasticity of the devices, enabling the emulation of neural behaviour and advanced information processing. The structure simulates excitatory synaptic activity through light stimulation and mimics lateral inhibition through electrical stimulation, effectively replicating the neural mechanisms of synaptic plasticity in processes such as Mach bands, contrast enhancement, and Hermann's grid. Leveraging these properties, we develop a lateral inhibition network for image recognition, achieving 97% accuracy-surpassing conventional networks at 93%. Additionally, through seamless integration with robotic arms, it can execute colour chip recognition on a machine cart, providing a promising strategy for the design of intelligent autonomous devices and bioinspired robots.

Pixel-programmable miniaturized optical arrays with large pixel count are essential for cutting-edge fields such as micro-displays, photonic chips, and light detection modules. In recent advances, a universal strategy with ultrahigh pixel count and highly flexible programmability remains lacking. Here we report a programmable optical nano-kirigami matrix with pixelated electromechanical reconfigurations. Deformable pixel arrays with high duty cycle and optical contrast are conceptually designed and experimentally realized based on a suspended turn-shaped nano-kirigami configuration. By employing the central plate to induce electrostatic force and the deformed arms to scatter incident light, switchable optical encryption and reconfigurable information display are demonstrated by programing the nano-kirigami matrices with a pitch size of only a few micrometers. Furthermore, line-level modulation based programmable information transmission and light projection are achieved by using a stripe-shaped addressable nano-kirigami matrix with 3.87 megapixels, showcasing an optical micro-array with large pixel count and flexible programmability. Our work enables the high visibility and precise addressability of freely controllable electromechanical arrays with massive pixels, which could greatly improve the practical applicability for miniaturized optical arrays and brings potential applications in micro-displays, photoelectronic chips, intelligent machine visions, hyperspectral image sensors, etc.

Underwater images suffer from wavelength dependent attenuation and multiple scattering, which often lead to severe color casts, veiling effects, and reduced contrast. These degradations weaken the stability of key vision modules such as feature detection, feature matching, edge extraction, and object recognition, and ultimately compromise applications including visual navigation, structural inspection, and environmental monitoring for underwater robots. To improve global color consistency while preserving local texture details, we propose a Physics-Guided Texture-Aware Fusion for Real-World Underwater Image Enhancement (GPRF-HPNet). In the preprocessing stage, a YCbCr domain attenuation map is exploited to guide color correction, followed by entropy driven dual histogram global contrast enhancement. The resulting intermediate images are further combined through gradient weighted wavelet fusion, which retains structural information and fine scale details. High frequency Gabor texture maps at four orientations, namely :0^circ:,45^circ:,90^circ: and 135°, are then constructed as an explicit detail prior. These maps feed a texture branch that runs in paralle with a base branch focusing on structure and color. A parallel residual fusion unit performs joint feature extraction on the two branches, learns adaptive weights, and produces fused feature representations, after which a lightweight decoder reconstructs the enhanced image. Extensive experiments on the Color-Check7, Test-C60 and UCCS datasets demonstrate that the proposed method achieves consistent gains on six metrics, including UIQM and UCIQE and delivers more reliable performance in downstream tasks such as geometric rotation estimation and edge detection, while demonstrating strong generalization across diverse underwater scenes.

Machine vision-based surface defect detection offers the advantages of being non-contact, non-destructive, and highly automated; consequently, it is widely applied across various industrial production processes. This article provides a brief overview of commonly used methods for surface defect detection, evaluation indicators for detection results, and key challenges currently faced. Defect detection methods are categorized into three types: traditional image processing methods, traditional machine learning methods, and deep learning methods. The core principles and representative studies of each method are reviewed, and their respective advantages and limitations are analyzed. We briefly describe the method for evaluating detection results, examine the few-shot learning problem encountered in practical applications, and provide an outlook on feasible pathways for addressing this issue in the future.

Biomimetic visual adaptation is crucial for machine vision to sustain robust perception over a wide luminance range. However, most existing adaptive optoelectronic devices rely on the joint regulation of external bias voltage and incident light intensity. Here, we present a trimodal organic active adaptation transistor (TM-OAAT) by integrating two bulk heterojunctions within the gate dielectric. This architecture enables synergistic modulation of photocapacitance enhancement and interfacial charge-trapping suppression. As a result, the device autonomously switches between scotopic, mesopic, and photopic vision modes without external gate bias modulation, covering a wide luminance range from moonlight to sunlight (10-2-106cd m-2). Imaging experiments and simulations demonstrate that the device effectively restores image features across all three adaptation modes, achieving recognition accuracy exceeding 97%. By achieving trimodal self-adaptation through illumination alone, this compact device provides a platform for low-power, wide-dynamic-range bio-inspired neuromorphic vision.

The digitalisation of pedigree accounting in the dairy cattle sector of the Kyrgyz Republic is crucial for improving animal husbandry efficiency and meeting increasing demands for food security and sustainable rural development. The main aim of the study was the development and piloting of software to automate the processes of identification, pedigree analysis, productivity and veterinary control of cattle. The methodology of the work included an analysis of existing solutions, the collection of functional requirements, the design of the database architecture and user interface, as well as the pilot implementation of the system in a real farm. The database structure implemented eight functional tableslinked through cascading foreign keys, ensuring the logical integrity of information and the automatic construction of pedigrees. The system interface is adapted to conditions of limited digital infrastructure and supports a multilingual environment, an offline mode and simple visual navigation. During the on-farm pilot study, a reduction in the time to register an animal by more than four times and a decrease in the number of errors were recorded. The system received positive user evaluations on the criteria of convenience, accessibility and accuracy. In addition, compatibility with International Committee for Animal Recording standards is ensured, which opens opportunities for international integration. The developed solution demonstrates resilience when scaling and can be adapted to other regions with similar infrastructure. The paper also provides a comparison with foreign and regional solutions, confirming the advantages of localisation, simplicity and compliance with national requirements. The study highlights the high practical significance of the proposed solution, offering a foundation for further scaling, including within national digitalisation programmes for agriculture.

Multi-rotor plant protection drones are extensively utilized in rice field operations. However, the airflow generated by the rotors not only disturbs the rice canopy but also affects the deposition of pesticide droplets. The investigation of the impact of rotor wind fields on the protection of rice crops has the potential to enhance the effectiveness of operations of this nature. The present study firstly acquired aerial imagery of plant protection drone operations via the utilization of aerial photography, and subsequently employed machine vision technology to investigate the disturbance patterns of rotor wind fields on rice canopies. Notably, there is currently no effective experimental method to observe the disturbance of rice canopies caused by the wind field generated by plant protection drone rotors, and machine vision processing is proven to be an effective approach, which is a key innovation of this study. This study innovatively adopts an integrated experimental-numerical approach, a distinct advancement over existing Unmanned Aerial Vehicle (UAV) spraying and airflow research that typically focuses on single experimental or simulation methods. Aerial photography and machine vision were used to explore canopy disturbance patterns, while reverse engineering combined with thrust tests and numerical simulations verified rotor model precision and analyzed droplet deposition. Core findings show that flight speeds of 3-4m/s enable effective overlap between canopy disturbance and droplet deposition zones, achieving optimal plant protection effects; higher speeds cause droplet drift and zone misalignment, reducing efficacy. In addition, this study explores the spray deposition of plant protection drones based on numerical simulation methods, and assists in judging the effect of plant protection operations through the coincidence of droplet deposition and canopy disturbance zones. This integrated approach and related findings provide a reliable technical basis for optimizing UAV flight parameters, and have certain reference significance for the technical research and development of plant protection drones and the setting of operation parameters.

The increasing presence of pesticide residues in vegetables poses a major threat to public health, and there is an urgent need to develop efficient, accurate, and scalable detection methods. Traditional analytical techniques such as Gas Chromatography (GC) and Liquid Chromatography–Mass Spectrometry (LC–MS) offer high sensitivity but are expensive, labor-intensive, and unsuitable for large-scale or real-time screening applications. Recent advances in spectroscopy, machine vision, and Machine Learning (ML) show promise; however, most existing models rely on handcrafted or shallow features and fail to capture the complex spatial, textural, and thermal variations associated with multi-residue contamination. In this direction, the present study proposes a data-driven hybrid deep learning framework for multi-residue risk classification in vegetables using thermal imaging. This framework integrates ΔT-based thermal image preprocessing, Gray-Level Co-occurrence Matrix (GLCM) texture descriptors, and deep feature embeddings extracted from InceptionV3, thereby forming a comprehensive hybrid feature vector. This fused representation is classified by a custom neural network trained with categorical focal loss to mitigate class imbalance and optimized using a cosine-decay learning rate to enhance convergence stability. Experimental evaluation on a custom thermal vegetable image dataset resulted in 84.97% validation accuracy and a loss of 0.0373, outperforming conventional Convolutional Neural Networks (CNNs) and other shallow classifiers. The model demonstrated good generalization with balanced precision and recall on contamination classes, supported by a well-converged training–validation performance and confusion matrix analysis. These results highlight the efficacy of this framework for non-destructive, real-time, and scalable pesticide contamination risk classification, pointing to its potential for deployment in automated food safety monitoring and smart agricultural inspection systems.