Robust Tomato Recognition for Robotic Harvesting Using Feature Images Fusion

Area of Science:

• Agricultural Robotics
• Computer Vision
• Image Processing

Background:

• Autonomous harvesting robots face challenges in fruit recognition due to complex agricultural environments.
• Illumination variations and overlapping fruits are key disturbances affecting robust fruit detection.
• Low-cost cameras require advanced algorithms for reliable fruit identification in uncontrolled settings.

Purpose of the Study:

• To develop a robust tomato recognition algorithm for autonomous harvesting robots.
• To address challenges posed by illumination and overlapping fruits in complex agricultural environments.
• To enable low-cost, reliable tomato detection using a standard camera.

Main Methods:

• Extraction of novel feature images: a*-component from L*a*b* color space and I-component from YIQ color space.
• Pixel-level image fusion using wavelet transformation to combine feature information.
• Adaptive thresholding for optimal segmentation and morphology operations for noise reduction.

Main Results:

• The proposed algorithm achieved 93% accuracy in recognizing target tomatoes from 200 samples.
• The method demonstrated effectiveness in handling illumination variations and overlapping fruits.
• Successful segmentation of tomatoes within a tree canopy using a low-cost camera.

Conclusions:

• The developed tomato recognition method is effective for robotic harvesting in uncontrolled environments.
• The combination of novel feature extraction and image fusion enhances recognition robustness.
• This low-cost approach offers a viable solution for autonomous agricultural applications.