An automatic visible-range video weed detection, segmentation and classification prototype in potato field.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Plant Science

Background:

• Uniform herbicide application leads to crop damage, environmental pollution, and increased costs.
• Site-specific spraying offers a sustainable solution to minimize herbicide use and environmental impact.

Purpose of the Study:

• To develop and evaluate a machine vision prototype for real-time identification and classification of Marfona potato plants and five common weed species.
• To enable site-specific weed management in potato fields, thereby reducing herbicide consumption and environmental pollution.

Main Methods:

• Utilized video processing and meta-heuristic classifiers for weed and crop identification.
• Extracted various features including color, texture (GLCM), spectral descriptors, moment invariants, and shape.
• Selected six key discriminant features for classification: saturation standard deviation (HSV), moment invariant difference, hue mean (HSI), area-to-length ratio, Cb component average (YCbCr), and in-phase component standard deviation (YIQ).

Main Results:

• Achieved a high classification accuracy of 98% correct classification rate (CCR) on the test set.
• Successfully differentiated Marfona potato plants from five weed varieties (mallow, purslane, lamb's quarters, rye, coklebur).
• The prototype demonstrated effective detection, segmentation, and classification of weeds in real-world field conditions at speeds up to 0.15 m/s.

Conclusions:

• The developed machine vision system is highly accurate in distinguishing potato plants from weeds.
• The prototype shows significant potential for implementing site-specific spraying strategies in agriculture.
• This technology can contribute to more sustainable and cost-effective weed management practices.