MACHINE VISION APPROACH TO DISCRIMINATE AND CLASSIFY NORMAL AND DISEASE-AFFECTED CROP SEEDS

Abstract

Every grower in the agriculture domain eventually observes the consequences of low-quality seeds, which include weak seedlings, moisture loss, inadequate stands, sluggish sprouting, and combined or ethnically compromised lots. The purpose is to have good quality and quantity crops from a marketing point of view. Selection of right seeds can make it possible. Till now, all this is handled manually, but machine vision approach can result better in this regard. Cotton, Wheat and Corn crops are widely cultivated in southern Punjab as well as throughout the world to produce various consumable and wearable items. Since seeds of these crops are evaluated in many areas with the aims of sowing and further processing, they should be identified in an expedited and accurate manner for the selection of a healthy (normal) seed. In this research activity, an affordable machine vision approach has been investigated to classify normal and diseased crop seeds. The experiments have been performed on the dataset gathered from Pakistan's Bahawalpur Agriculture Research Center. The domain specialists sorted plants that were healthy from those that were afflicted by disease. Several of the best forecast models for classifying seeds as normal and unhealthy for all three crops have been obtained through the investigation of feature sets, feature models, and machine learning classifiers. To adjust the algorithm's variables, tests as well as training sets have been utilized with the K-fold cross-validation method. Resultantly, fast and accurate operations have been adopted to serve agricultural industries. For quantitative assessment of the research work, standard evaluation parameters, accuracy, specificity and sensitivity have been used. The developed system achieved an overall accuracy for cotton as (83.8%), for wheat as (100%) and for corn as (84.0%).