Optimization of color index and threshold segmentation in weed recognition

The impact of potential classification error on machine-vision weed recognition has stimulated research into new methods of optimizing segmentation. The color index and threshold for weed image segmentation are transformed into the segmentation surface in RGB color space. The evaluating method of segmentation error was established with Bayes formula, and color indexes were optimized and threshold parameter was processed via genetic algorithm. Optimal segmentation surface is -149R+218G-73B=127. With a comparison of the experimental results between Excess-Green method and new segmentation surface method, the segmentation noise of the new method is lower than the former, the average probability of segmentation error decreases from 3.90% to 2.33%. It is more propitious to the extraction of shape feature in the next classification operation.