Image-Based Plant Disease Identification by Deep Learning Meta-Architectures.

Area of Science:

• Agricultural Science
• Computer Vision
• Deep Learning

Background:

• Accurate plant disease identification is crucial for crop monitoring.
• Computer vision and deep learning offer advanced solutions for agricultural challenges.

Purpose of the Study:

• To perform localization and classification of plant diseases using deep learning meta-architectures.
• To evaluate and optimize the performance of these models for disease detection.

Main Methods:

• Applied three deep learning meta-architectures: Single Shot MultiBox Detector (SSD), Faster Region-based Convolutional Neural Network (RCNN), and Region-based Fully Convolutional Networks (RFCN).
• Utilized the TensorFlow object detection framework for model training and testing on a controlled dataset.
• Optimized the best-performing architecture using state-of-the-art deep learning optimizers to improve mean average precision (mAP).

Main Results:

• The Single Shot MultiBox Detector (SSD) model, trained with the Adam optimizer, achieved the highest mean average precision (mAP) of 73.07%.
• Successfully identified 26 different types of diseased leaves and 12 types of healthy leaves within a single framework.
• Demonstrated the effectiveness of deep learning models in complex plant disease localization and classification tasks.

Conclusions:

• The proposed deep learning methodology, particularly the SSD model with Adam optimizer, shows high efficacy for plant disease detection.
• The developed framework can be adapted for various agricultural applications and real-time disease monitoring.
• The generated model weights offer potential for future reuse in diverse environmental conditions.