Comparative Analysis of Detectors and Feature Descriptors for Multispectral Image Matching in Rice Crops

Area of Science:

• Agricultural Science
• Computer Vision
• Image Processing

Background:

• Precision agriculture leverages machine vision and image processing for crop management.
• Combining visible spectrum (VIS) and near-infrared (NIR) images aids crop identification.
• Image matching between different sensors presents a significant challenge due to camera variations and distortions.

Purpose of the Study:

• To compare the performance of various feature descriptors and detectors for image matching in rice crops.
• To evaluate different spectral bands (RGB, CIE L*a*b*, NIR) for matching accuracy.
• To identify the optimal algorithm for robust crop identification in precision agriculture.

Main Methods:

• Acquisition of VIS and NIR images from different cameras across 20 rice crop scenes.
• Extraction of RGB and CIE L*a*b* channels from VIS images for comparison with NIR images.
• Implementation and evaluation of feature detection (FAST) and description (BRISK, SURF, SIFT, ORB, KAZE, AKAZE, BRIEF, FREAK) algorithms using Python and OpenCV.

Main Results:

• The green channel consistently yielded the highest number of correct matches across all tested methods.
• The combination of the FAST feature detector and the BRISK descriptor demonstrated superior performance.
• This FAST-BRISK combination achieved the best balance between processing time and accuracy in matching VIS and NIR images.

Conclusions:

• The FAST feature detector paired with the BRISK descriptor is highly effective for image matching in precision agriculture.
• Utilizing the green channel from VIS images alongside NIR data significantly improves crop identification accuracy.
• This optimized approach enhances the reliability of machine vision systems for agricultural applications.