Title : Deep learning based automated In-depth quality inspection for fruits & vegetables
Abstract:
Fruits and vegetables are an essential part of everyone's life and their quality is a key factor in consumer buying decisions. Because of this, quality inspection (QI) by retailers while sourcing becomes essential. Performing QI as per defined standards at scale and with consistency is a challenging task. This paper presents a method to automate the inspection (visual) of almost all the fruits and the vegetables following any set of defined standards. The QI (visual) problem is formulated into an object detection & classification (ODC) problem and our algorithm, a convolutional deep neural network, was trained for several types of defined defects and their severity. Another QI parameter, produce object size, was estimated for individual objects by first segmenting it, then creating contours and processing it to find distance between two farthest points. On evaluating the respective ODC model on cucumber and lime test set, the models achieve accuracy of 75% on cucumber and 69% on lime. Similar results were obtained for other produce. On size estimation, the method achieves the accuracy of 91.56% and precision of 97.94%.
