Proven Case Studies

Challenge

Onion sorting relies on manual labor, with anomalies affecting quality and yield. Conventional image recognition lacked accuracy in real-world conditions.

Solution

Developed a feature-adaptive anomaly detection method, validated in both lab and real-world operational environments.

Impact

Achieved 97% accuracy in lab and 86% in real-world operation.

Challenge

Konjac tubers have irregular shapes with frequent occlusion, making size and weight estimation inaccurate with conventional methods.

Solution

Developed a size and weight estimation method that handles occlusion, enabling accurate estimation even for partially hidden tubers.

Impact

Reduced size error by over 50% and weight error by 72%.

Challenge

Harvest timing depends on visual color assessment, making it difficult for inexperienced farmers. An objective criterion was needed.

Solution

Developed a color estimation model using Vision Transformer to accurately determine harvest readiness from images.

Impact

Achieved 97.2% accuracy in color estimation, enabling objective harvest timing decisions.

Challenge

Thrips are tiny pests, and conventional image recognition achieved only 66.5% detection accuracy.

Solution

Developed a detection method enhanced with super-resolution, amplifying tiny pest features to significantly improve accuracy.

Impact

Improved detection accuracy from 66.5% to 89.7% (+23.2 points).

Challenge

Grape grading relies on expert visual assessment, leading to inconsistent standards and labor shortages.

Solution

Developed a deep learning grading system fusing multi-view imaging with IoT sensor data.

Impact

Achieved up to 85.7% grading accuracy.

Challenge

Grape berry thinning relies on skilled farmers' experience and intuition, making it difficult for newcomers and unskilled workers to learn.

Solution

Developed a system combining deep learning berry detection with AR glasses to display real-time thinning guidance.

Impact

Unskilled farmers achieved 8.18% higher quality scores than skilled farmers when using the system.

Challenge

Grape berry thinning is highly manual labor, with severe workforce shortages and physical burden demanding automation.

Solution

Developed an automated thinning system combining berry detection AI with a robotic arm, validated in both indoor and field environments.

Impact

Achieved 97% indoor and 90% field success rate for automated thinning.