Automated Millet Classification Using Transfer Learning and Computer Vision

ABSTRACT The precise identification and classification of millet grains are essential for improving the efficiency of the agricultural processing systems. While there are sorting machines available for this purpose, the equipment is quite expensive and not easily accessible for improving the efficiency of the agricultural processing systems. This study proposes an automatic classification framework using computer vision and transfer learning techniques to classify the millet grains into 14 classes, consisting of major millets like sorghum, pearl, and finger millets, and minor millets like processed and unprocessed millets, depending on the presence of the hull. Microscopic images are employed to capture the minute differences in the millet grains. Four different models of convolutional neural networks were considered for this purpose, and the best performance was obtained with the MobileNet model. Fine‐tuning the model further improved the accuracy of the classification to 99.3%.