Fruit Quality Determination using Image Processing and Deep Learning

Abstract

A considerably high amount of fruit produced is wasted due to improper management and utilization during harvesting, storing, transporting, and in the food processing industry. Fruit will get rotten easily if not stored properly due to bacteria accumulation. It is known to all that rotten or defective fruits are harmful to health. It may damage the fresh fruits which are in surface contact with the rotten fruits in the inventory. These rotten fruits should be detected and sorted as early as possible. The problem that comes across in manual checking by humans is less uniformity and accuracy as the manual examination by humans’ eyes will consume time and energy. This research proposes a method involving the deep learning technique which is CNN (Convolutional Neural Networks) for feature extraction and classification of rotten fruits. It is one of the applications of image classification problems. This approach uses an RGB channel image of the fruit under examination. The image will be evaluated by the trained model as fresh if the percentage of rotten part detected is under the threshold value. The types of fruits that will be identified and classified in this paper are apple, banana and orange. Transfer learning technique is used, which minimizes training time and resources and aids to achieve higher accuracy. The dataset is divided into two parts, for (70%) training and (30%) validation. The raw image set used for training is first pre- processed and then fed into the model. The validation accuracy obtained in this paper is 98.47%. The total duration of the training stage is 210.37 minutes. Hence, the required time to classify a single fruit image is approximately 0.2 second. Our model can be adopted by industries closely related to the fruit cultivation and retailing or processing chain for automatic fruit identification and classifications in the future.