Scholarlink

<p>In recent times, global food security has been threatened by plant leaf disease, which reduces crop yield and affects millions of people’s livelihoods. Due to the variations in climatic conditions such as rainfall, temperature, and humidity, plants are continuously exposed to harmful pathogens, including viruses, bacteria, and fungi, which result in poor crop yield. To resolve this challenge, this research proposes a&#xa0;novel Depthwise Squeeze-based Residual Recurrent Autoencoder framework for the accurate prediction of plant leaf disease using multimodal data. In this study, the Depth Dilated-based Visual Geometry Group&#xa0;16 is utilized to capture the spatial features from plant leaf images, and the model deploys a&#xa0;Depthwise Separable Convolution and Dilated Convolution for enhancing the fine-grained feature extraction while minimizing computational complexity. The Bidirectional Swish Recurrent Unit is implemented to extract temporal patterns from diverse environmental sensor data, in which the swish activation function boosted the non-linear modeling as well as gradient flow. In addition, the Principal Component Analysis is implied to reduce the dimensionality of the extracted features and computational load at the same time, preserving the significant disease-relevant details. Moreover, the strategy exploits a&#xa0;Residual Squeeze Variational Autoencoder for efficient plant leaf disease prediction by combining a&#xa0;Gaussian Residual Module and a&#xa0;Squeeze-and-Excitation mechanism that rendered effective latent representations and pointed out the crucial features. The experimental validation revealed that the Depthwise Squeeze-based Residual Recurrent Autoencoder scheme achieves a&#xa0;high accuracy of 98.87%, a&#xa0;high precision of 98.71%, a&#xa0;minimal false positive rate of 1.44%, and a&#xa0;minimal computational time of 0.045 sec. These outcomes guaranteed its robustness, accuracy, as well as suitability for real-time agricultural applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Depthwise Squeeze-Based Residual Recurrent Autoencoder for Accurate Plant Leaf Disease Prediction Using Multimodal Data

  • 期刊论文

  • 发表日期：

    2026年2月13日

查看全文
  • P. Jagadeesan,
  • M. Vedaraj

摘要

In recent times, global food security has been threatened by plant leaf disease, which reduces crop yield and affects millions of people’s livelihoods. Due to the variations in climatic conditions such as rainfall, temperature, and humidity, plants are continuously exposed to harmful pathogens, including viruses, bacteria, and fungi, which result in poor crop yield. To resolve this challenge, this research proposes a novel Depthwise Squeeze-based Residual Recurrent Autoencoder framework for the accurate prediction of plant leaf disease using multimodal data. In this study, the Depth Dilated-based Visual Geometry Group 16 is utilized to capture the spatial features from plant leaf images, and the model deploys a Depthwise Separable Convolution and Dilated Convolution for enhancing the fine-grained feature extraction while minimizing computational complexity. The Bidirectional Swish Recurrent Unit is implemented to extract temporal patterns from diverse environmental sensor data, in which the swish activation function boosted the non-linear modeling as well as gradient flow. In addition, the Principal Component Analysis is implied to reduce the dimensionality of the extracted features and computational load at the same time, preserving the significant disease-relevant details. Moreover, the strategy exploits a Residual Squeeze Variational Autoencoder for efficient plant leaf disease prediction by combining a Gaussian Residual Module and a Squeeze-and-Excitation mechanism that rendered effective latent representations and pointed out the crucial features. The experimental validation revealed that the Depthwise Squeeze-based Residual Recurrent Autoencoder scheme achieves a high accuracy of 98.87%, a high precision of 98.71%, a minimal false positive rate of 1.44%, and a minimal computational time of 0.045 sec. These outcomes guaranteed its robustness, accuracy, as well as suitability for real-time agricultural applications.