ANALYTICAL IMPACT OF PRE-PROCESSING AND AUGMENTATION ON WHEAT LEAF IMAGES

Abstract

Wheat, a staple crop for a significant chunk of the universal inhabitants, suffers substantial yield deprivation due to various diseases. Wheat leaf disease identification play noteworthy part in early detection and control the disease. The effective and efficient wheat leaf disease detection depends on clean and good quality wheat leaf image data set. In this study we assess the impact of pre-processing and augmentation approaches on wheat leaf images in convolutional neural architecture. We also analyse the data enhancement technique such as data augmentation and parameter optimization. All experimental work were validate using a k- layer (here k is 4) cross-validation to check the stability of model. K-layer (set) approach helps us to build a more real time based model. Validation accuracy and test accuracy shown here of all models used in our experimental work. The result suggest the most CNN performs better when we use original plus pre-process images. The impact of data augmentation on wheat leaf images also enhance the model accuracy.  In terms of validation and test, both cases accuracy increases Alex Net and Inception V3 CNN models shows the enhanced accuracy in this paper.