Automated Vision‐Based Arecanut Defect Detection and Classification Using CNN and IRB360 Industrial Robot

Arecanut is a major crop in India, holding economic and cultural significance. However, significant losses occur due to fruits falling and manual collection of these fallen nuts is exhausting and time‐consuming. Existing defect detection methods often lack adaptability and accuracy for diverse object types, complicating quality control. This study aims to address these challenges by proposing an automated system to identify and collect good quality fallen arecanuts, while discarding the defective ones. The system utilizes a vision‐guided robotic arm and deep learning techniques. A real‐time dataset was built by collecting and preprocessing arecanut images. Convolutional neural networks (CNN) and traditional machine learning methods such as support vector machine (SVM), random forest (RF) were evaluated and compared for defect detection and classification. The system integrates force‐based classification with real‐time object handling. A PicArecanut (PA) algorithm using RAPID programming was developed to control the robotic pick‐and‐place operation. CNN outperformed SVM and RF, achieving a classification accuracy of 86.44%, compared to 84% and 81%, respectively. The robotic system, using PickMaster, demonstrated high precision (0.94) and recall (0.93) in classifying arecanuts. This demonstrates CNN's superior ability to generalize across defect types. The deep learning‐powered robotic system offers a scalable, efficient, and cost‐effective solution for arecanut collection, reducing manual labor and minimizing losses.