Smarter Downsampling and Attention Help AI Read Faces More Accurately
Researchers in China have improved facial expression recognition by redesigning how a ConvNeXt network downsamples images and integrates spatial-channel attention, ...
Tag: Computer Vision
AI Transformer Hits 98% Accuracy Spotting Jute Pests Before They Devastate Crops
Researchers in India have developed a hierarchical feature fusion transformer that identifies seventeen jute pest species with 98.43 percent accuracy, ...
AI That Shows Its Work: Explainable Deep Learning Spots Crop Pests in Real Time
Researchers have unveiled an explainable deep learning framework that detects insect crop pests in real time with up to 98 ...
Traffic-Light Fluorescent Probe Delivers Rapid Mercury Warnings via Deep Eutectic Chemistry and AI Vision
Researchers have built a deep eutectic fluorescent composite paired with YOLOv8 computer vision that detects mercury ions at 60 nanomolar ...
Drone Learns to Chase Moving Targets by Fusing Vision AI With Reinforcement Learning
Researchers in China have built a drone trajectory generation system that combines YOLOv8 detection, LSTM motion prediction, and Q-learning-optimized nonlinear ...
AI Fails at Reading Plant Cells Under the Microscope, New Benchmark Reveals
University of Queensland researchers have created PlantMicro, a benchmarking dataset showing that vision language models perform at barely above random ...
AI Turns an Ordinary Tablet Camera Into a Precise Balance Test Scorer
A camera-based AI system running on an ordinary tablet scored the Balance Error Scoring System more consistently than real-time human ...
New AI Rebuilds Heavily Damaged Faces While Keeping Identities Intact
Researchers have unveiled a generative adversarial network that rebuilds faces obscured by up to 70 percent occlusion while preserving the ...
Deep learning model watches warehouse workers to spot dangerous moves in real time
A deep learning pipeline combining YOLOv5 detection, skeleton extraction and neural network classification can recognize falls, collisions and shelf climbing ...
AI Network Spots Hidden Wind Turbine Blade Damage From Drone Photos
Researchers have developed CHS-Net, a hybrid deep learning architecture that sets a new benchmark for segmenting cracks, erosion, and delamination ...
Enhanced AI Vision System Teaches Greenhouse Robots to Actually Pick the Fruit
A new RGB-D perception-to-action framework called E-YOLOv11-RGBD raised robotic greenhouse harvesting success from 51.7 to 92.5 percent by converting AI ...
Graph Neural Networks Push Diabetic Retinopathy Detection Toward Near-Perfect Accuracy
Researchers in India have built a hybrid graph neural network framework that converts retinal images into graph structures, achieving up ...
AI Watches the Herd: Vision System Tracks Every Cow in a Crowded Barn
A hierarchical computer vision framework combining object detection, SAM2 segmentation, and multi-object tracking enables contactless, individual-level dairy cow behavior monitoring ...
Hybrid AI Turns Network Traffic Into Images to Catch Cyberattacks
Researchers in Algeria have combined convolutional neural networks with a Swin Transformer to detect network intrusions with up to 99.69 ...
Lightweight Transformer Brings Real-Time Road Garbage Detection to Edge Devices
Researchers in China have developed a lightweight transformer model with a novel cross-axis attention module and a feature memory component ...
AI Watches the Whole Flock: Trajectory-Free Anomaly Detection for Dense Hen Houses
Researchers have developed a trajectory-free, semi-supervised anomaly detection framework that monitors entire hen flocks through statistical activity and density cues, ...
Dual-Stream AI Combines Compression Clues and Deep Vision to Expose Doctored Images
Researchers have developed a dual-stream convolutional neural network that fuses RGB imagery with vectorized Error Level Analysis and CBAM attention ...
Deep Learning Learns to Tell Rice Varieties Apart, One Grain at a Time
Researchers in Türkiye combined transfer learning with MobileNetV2, Xception, and classical classifiers to classify five rice varieties with up to ...
New Open-Source Tool Fixes the Hidden Refraction Errors That Warp Underwater 3D Vision
Researchers have released AquaCal, an open-source Python library that corrects the depth-dependent errors caused by light refraction in multi-camera calibration ...
New AI Framework Reads Gene Maps Like Images to Reveal Hidden Cell States
A new computational framework called STARIT converts the spatial positions of mRNA molecules inside cells into image tensors, allowing deep ...
