140 likes | 285 Views
Image Processing for Plant Recognition in Mechanical Weeding Applications. Fergus Browne ECE 09427091. Project Aim. Develop algorithm to detect plants and ignore weeds using OpenCV Implement algorithm on Smartphone or other suitable platform. Algorithm 1. Colour Detection
E N D
Image Processing for Plant Recognition in Mechanical Weeding Applications Fergus Browne ECE 09427091
Project Aim • Develop algorithm to detect plants and ignore weeds using OpenCV • Implement algorithm on Smartphone or other suitable platform
Algorithm 1 Colour Detection • Finding RGB range of colours for plants • Ignore everything outside of range Ideal Conditions • Little to no weed coverage • Average sunlight
Results • Blurring makes shapes inaccurate • Weeds can have very similar colours to plants • 2/12(16.66%) images had successful results
Algorithm 2 Contours and Hough Circles • Search for contours above a specific threshold • Examine contours found, surrounding those likely to be part of a circle in Hough Circle Key Points • Different images require different threshold values • Not all plants have circular shapes
Results • 3/12 (25%) of images had successful results • Large range of threshold values required(manual input!)
Algorithm 3 Combining Colour Detection with Contours • Colour detection narrows down areas of interest • Ideally would reduce range of threshold value across different images and increase number of useful results
Results • 2/12 (16.66%) images provide useful results. • Threshold range reduced from 151 to 238 (87) and new range ran from 184 to 223(39) • Does not improve on previous algorithm, less successful results
Algorithm 4 Template Matching • Search for image within larger image • Examine template at each point in image, calculate matching coefficient • Draw rectangle at best point
Conclussions • Colour detection unable to narrow down areas of interest sufficiently • Template Matching proving very promising (5/5 successful results)
Next Steps • Improve on Template Matching by allowing it to detect a variety of plants • Create reference kernels representing plants and search for them • Implement the final algorithm on a suitable platform, such as a Smartphone