PlantNet: Adaptive DenseNet with Attention-Based Capsule Network for Classifying Plant Diseases with Segmentation Procedures in Agricultural Fields

Abstract

Due to the rapid growth of population and increased demand of food, agriculture plays an essential role worldwide. Here, the quality as well as the quantity of the agricultural production is greatly affected by plant disease. Recently, various advancements in plant disease have been greatly identified by considering deep learning models to provide promising outcomes. The disease symptoms appear on the leaves and are easily identified through recent developments in computer vision and deep learning techniques. Classifying the plant disease at the initial stage becomes a tedious process that impacts the performance of traditional models. Also, conventional techniques do not work properly in large image datasets, affecting the model’s efficiency while focusing on a wide variety of conditions. Hence, this paper aims to implement an efficient plant disease detection and classification approach with deep learning methods. Initially, the required data from publicly available sources are collected and provided to the image segmentation phase. Here, the developed Dilated Dense R2Unet (DD-R2Unet) technique is utilized to segment the appropriate regions in the collected plant disease-affected images. Further, the segmented images are given to the plant disease classification model. In this phase, the developed framework utilized the Adaptive DenseNet with Attention-based Capsule Network (Ada-D-ACapsNet). Later, several hyperparameters in the implemented Ada-D-ACapsNet are tuned using a novel optimization technique named Enhanced Controlling Parameter-based Humboldt Squid Optimization Algorithm (ECP-HSOA) to enhance the plant disease classification efficiency. Finally, the plant disease classified outcomes are attained from the developed Ada-D-ACapsNet. Further, various performance analysis is executed in the implemented plant disease detection and classification approach over classical models. During the performance evaluation, the developed model shows 93.88%, 99.45%, and 93.89% in terms of accuracy, specificity and Negative Predictive Value (NPV). This performance enhancement facilitates accurate plant disease classification promptly.

Plant phenotyping relevance

植物病害の症状を画像からセグメンテーションし、病害状態を分類する画像ベースの表現型推定手法が研究の中心であり、技術比較による性能評価も行っている。

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.