Categorization and Grading of Spices Using Deep Learning

Detection of spices from images and recognition based on image processing is a popular research topic. Food is the most vital aspect of human life. Every food consumed on a daily basis contains a variety of spices that make it delicious and flavorful. People are especially concerned about their health. The spices used in meals play a vital role in the prevention of dietary, obesity, and other such issues. The aim of this project is to create an automatic system for detecting and recognising Indian spices on different images, so that dieticians may properly analyze nutrition and other types of health dangers. Color and texture data were primarily used in the recognition method for a better categorization outcome. This approach has been evaluated on a variety of spices. The different spices were classified using a Convolution Neural Network (CNN). The proposed system involves the CNN model for categorization. The spices dataset is generated by collecting images from the internet and creating more images for training by using data augmentation for 4 categories. The spices images contain 640 as training data and another 128 images taken separately for testing data. For obtaining an optimum model with increased classification accuracy different combinations of number of hidden layers and epochs are analyzed. The overall network performance losses for various cases are also observed. Experimental results produced the best test accuracy of 91.14% and the best training accuracy of 97.19%.