Validation images were classified into two categories based on a predetermined brightness using a color histogram. CHT was initially run on the validation images to detect spherical objects, and a maximum squared patch was taken from the inside of each circle. After dividing the image into 20 x 20 smaller patches, the SVM classifier was run to classify each of them, and it was determined if the image was fruit or leaf.
A false positive algorithm was then run, and based on majority vote, false positive fruit were removed. In the secondary method of false positive detection, a SIFT keypoint detection algorithm was run on each larger patch, and the number of keypoints was counted on each larger patch. Since the surface of a citrus fruit is relatively smooth, it would have a smaller number of corners, thus, the number of detected keypoints would be lesser for the fruit surface compared to the leaf surface.
Using the photos of the fruit and the CHT, texture classification with a support vector machine, and keypoints by scale invariant feature transform algorithms, Lee and his team found 80% of immature fruit in the research grove on the Gainesville, Florida campus. Major challenges identified in the study were uneven illumination, partial occlusion, and color similarity. Despite an 80% success rate, the number of false positives identified raised an issue, so further research is necessary to increase the accuracy of the identification process.
View the University of Florida press release.
View the research paper.
Also check out:
MACHINE VISION SOFTWARE: Vision algorithms added to graphical programming software
Applying algorithms for machine vision
IMAGE-PROCESSING SOFTWARE: Scientists advance object recognition and scene-understanding algorithms
Share your vision-related news by contacting James Carroll, Senior Web Editor, Vision Systems Design
To receive news like this in your inbox, click here.
Join our LinkedIn group | Like us on Facebook | Follow us on Twitter | Check us out on Google +
Page 1 | Page 2
