Abstract
For accurate segmentation of lettuce canopy images, dealing with uneven illumination and background interference, hyperspectral imaging technology was applied to capture images of lettuce from the rosette to nodule stages. The spectral ratio method was used to select the characteristic wavelengths, and the characteristic wavelength images were denoised and image fused before being processed by filtering and threshold segmentation. To verify the accuracy of this segmentation method, the manual segmentation method and the segmentation method used in this study were compared, and the area overlap degree (AOM) and misclassification rate (ME) were used as criteria to evaluate the segmentation results. The results showed that the segmentation effect was the best when 553.8 nm, 702.5 nm and 731.3 nm were selected as the characteristic wavelengths of lettuce for the spectral ratio method, with an AOM of 0.9526 and an ME of 0.0477. Both have a variance of less than 0.01 and have the best stability. Hyperspectral imaging technology combined with multi-wavelength image and multi-threshold segmentation can achieve accurate segmentation of lettuce canopy images.