Comparative Analysis of Deep Convolutional Neural Network for Accurate Identification of Foreign Objects in Rice Grains

The presence of foreign objects in rice collections is known to pose significant concerns for both food safety and product quality. Neglecting to address these issues can result in a deterioration of rice quality, a decline in its economic value, and a loss of consumer trust and confidence. Therefore, this research introduced a novel model for the recognition of natural and manufactured foreign objects through the use of Deep Convolutional Neural Network (DCNN). The model focused on analyzing images of rice for easy identification and classification. DCNN model classified the images into six groups based on different types, namely stone, paddy, fragment rice, broken, broken -yellow, and black -red. These classifications were determined using three pre -trained models, such as ResNet50, VGG16, and MobileNetV2. This research also used comparative techniques for multi -class classifications, combining multiple machine learning with digital image processing measurements and comparative performance. Despite using a cross -validation model, the experiment proved that using a pre -trained technique with transfer learning produced higher accuracy than a typical machine learning model. The most accurate prediction was made by VGG16 using a transfer learning of 97% compared to a random forest (RF) value of 94% without cross -validation with 5 K -fold and 10 K -fold cross -validations.