An Artificially Intelligent Framework for Plant Health Monitoring

Abstract

Plants are cultivated and consumed all over the world. They are highly nutritious and are rich in vitamins, minerals etc. However, most plants are vulnerable to biotic and abiotic diseases, which limits the yield. So, it is essential to detect and handle these diseases at the earliest to get an ample amount of produce. Computers and digital devices make this process much more effortless than manual human intervention. The general template followed by researchers is first segmenting out the diseased lesions from the leaves and then applying machine learning classifiers to differentiate between the diseases. Our proposed solution involves classification followed by lesion isolation and quantification. The techniques are automatic and require no human intervention in the segmentation steps. Additionally, most of the classification work is done on specific plant and disease combinations like cherry powdery mildew, apple rust etc., which requires retraining the classifier in the case of the introduction of a new disease-leaf pair. We tried to solve this limitation by classifying the leaf images based on the disease type, not disease-leaf pairs, as most diseases have similar infection patterns in different plants. Our study included powdery mildew, rust, and bacterial spot diseases. A classification model has been developed which classifies whether a leaf is suffering from powdery mildew, rust, bacterial spot or is healthy. A remarkable accuracy of 99.09% was observed on the test dataset. Moreover, the detection techniques are robust to various lighting conditions, leaf color patterns and symptom patterns

Plant phenotyping relevance

葉画像から病斑を自動分離・定量し、植物の病害状態を分類する画像解析フレームワークの開発であり、植物フェノタイピング手法が中心です。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.