Citrus Fruits Classification and Evaluation using Deep Convolution Neural Networks: An Input Layer Resizing Approach

Plant diseases are one of the causes of financial deficits in the agriculture farming industry around the world. It is the most important factor since it reduces both the potential and the quality of the growing crops. Fruits are one of the most important sources of nutrients in plants; although, there's a diverse range of diseases that affect the quality and development of fruits. Machine vision approaches have been developed to deal with these challenges these approaches are very useful in that they not only identify diseases/infections early but also tag them. In this paper, a deep CNN-based approach for disease classification of citrus fruits into healthy and infected classes is proposed. To improve the quality of image samples, the dataset is pre-processed with normalization. CNN model is implemented on the citrus fruits dataset with different sizes of input layers such as 64×64,100×100and 256×256.it is observed that the model having the size of 64×64 and 256×25 produces the same accuracy but the size of the model with 64×64 is significantly reduced by the factor of 37.65% on our collected dataset. The maximum accuracy of 96% is obtained with a model of input layer size of 100×100 with the 38 MB size.