Damaged Apple Detection Using Artificial Intelligence

Abstract

The field of mechanical engineering is evolving with latest technologies such as artificial intelligence. the blend of AI technologies such as deep convolutional neural network (DCNN), convolutional neural network (CNN), artificial neural network (ANN) which contributes more to control the process parameters, process planning, machining, quality control and optimization for a better product or system. The implementation of AI in mechanical engineering applications results in minimizing the rejection of machine components which helps the whole process to be economical with better quality outputs. Considering the stiff competition among the manufacturers in the market, increasing the production rate while maintaining stringent quality control is a big challenge. In this perspective, artificial intelligence is gaining popularity in production lines to maintain a high quality for the products. A CNN is a deep learning algorithm, that is analogous to that the connectivity pattern of neurons in the human brain, has become popular and effective to image classification problems recently. It takes in the image of the object and assigns importance to various aspects/objects in the image so as to differentiate one from the other. In fruit-sorting process, manual classification is time-consuming, expensive, and requires experienced experts whose availability is often limited. To address these issues, various machine learning algorithms have been proposed to support the automated classification of fruits. In this paper, to classify “regular apples” and “damaged apples”, deep learning algorithm is applied. The pre-trained, deep learning models namely, VGG 16, ResNet50, Inceptionv3, Mobilenet_v2 along with a basic sequential convolutional model are applied to differentiate the damaged apples from regular ones and their performance variation is also analyzed. For this work, the data set containing damaged and regular apples was garnered from various local stores and farms. The data set consisted of 400 color images of both regular and damaged apples. Though the number of samples is smaller, the above-mentioned deep learning models demonstrated to overcome this deficit. For the training of model, 80% of the total sample (280) images were utilized while 20% and 10% of the sample (80 & 40) were applied for the validation and testing the model. The results show more than 90% accuracy for all the models except ResNet 50. The performance of these models can be improved even further by increasing the size of data set by adding more fruit images through better training of the models. Our experimental study demonstrates the application of artificial intelligence through four different transfer learning techniques works well for deep neural network-based fruit classification. It minimizes the labor and human errors involved in the fruit-sorting process which results in saving money and time.