Identification of Diseased Apple Fruit using 3D and 2D Convolutional Neural Network for Improving Accuracy

Abstract

To identify the diseased apple fruits using 2D and 3D convolution neural network. Materials and Methods: Kaggle dataset is used to characterize the images, dataset contains 100 images. In training 50 images, (Healthy 25 and Diseased 25) and testing 50 images (Healthy 25 and Diseased 25) are used and the algorithm is iterated 25 times (N) for all samples. The images are tested using two machine learning algorithms 2D Con-volutional Neural Network and novel 3D Convolutional Neural Network. The samples are calculated using Gpower with pretest power of 80confidence interval of 95Novel 3D Convolutional Neural Network achieved classification accuracy 84.64 % and 2D Convolutional neural network achieved accuracy of 70.92 % with 2 tailed significance value of 0.001 (P!0.05). Conclusion: Novel 3D Convolution Neural Network appears to be significantly better than the 2D convolutional neural network in terms of accuracy for identifying diseased fruits.