Visual Categorization of Fruit disease using Fine-Grained Stacking Ensemble Learning

Abstract

Marssonina blotch infect the entire tree and black rot creates the possibility for fungicides to infect the neighborhood fruits thereby creating the entire fruit basket infected. Marssonina blotch and black rot are the most commonly occurring infections on fruits. Scab is discovered as the other category of fruit infection caused by a fungus infecting even leaves that cause cracks and immature stage of fruits and leaves. The preprocessing steps involve the process of segmentation, filtration and haar cascade defect training using the training data set consisting of 2500 defect records on apple, strawberry, grapefruit, cucumber, kiwi, lime, mango and guava. Deep learning and Stacking ensemble learning is used to diagnose the infections on fruits by using the CNN algorithm to create fine grained chunk visuals for disease prediction. The system is constructed to identify color of the fruit so that the system is able to predict the fruit image on all the possible colors for example on apple both red and green color apples are trained. The overall accuracy of prediction recorded by the proposed work is 97.8% which proves to hit the required efficiency of the diagnosis system to prevent the infection on fruits and increase the productivity and marketing strategy. Blotch prediction accuracy is observed as 97.5%, scab as 95.89 %and rot prediction accuracy is recorded to 98.2%.

Plant phenotyping relevance

果実画像から病害状態を分類・診断する画像解析手法が研究の中心であり、植物の病害表現型を直接推定しているため。

Code and data availability

The preprint describes fruit disease classification using a dataset of fruit images and CNN/stacking ensemble models, but provides no public dataset link, repository, code availability statement, or trained model release. The only supplementary file is an author biography. No paper-specific public asset is available.

No evidence-backed public reproduction asset is currently recorded.