Enhanced Field-Based Detection of Potato Blight in Complex Backgrounds Using Deep Learning.

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Manual detection of potato blight is time-consuming and requires expertise.
• Timely identification of blight disease is crucial for effective crop management.
• Automated systems can overcome limitations of manual disease diagnosis.

Purpose of the Study:

• To develop an automated system for detecting blight disease in potato leaves.
• To utilize Mask Region-based convolutional neural network (Mask R-CNN) for blight detection.
• To evaluate the performance of different color spaces in improving detection accuracy.

Main Methods:

• Implemented Mask R-CNN with a residual network backbone for detecting blight patches.
• Trained the model on 1423 field images with over 6200 labeled patches.
• Converted RGB images to HSL, HSV, LAB, XYZ, and YCrCb color spaces for comparative analysis.

Main Results:

• The Mask R-CNN model successfully differentiated diseased leaf patches from background soil.
• LAB color space model achieved 81.4% mean average precision; HSL achieved 56.9% mean average recall.
• Overall precision reached 98% on images with complex field backgrounds.

Conclusions:

• Automated blight detection using Mask R-CNN is feasible and effective in field conditions.
• Color space transformation can enhance the performance of blight detection systems.
• The developed system offers a promising tool for early blight disease management in potato cultivation.