Automatic detection of plant diseases; utilizing an unsupervised cascaded design

This paper describes a machine vision framework that determines the ocular manifestation of plant diseases by analyzing the images in CIELab color space. The major contribution of our article is to build up a procedure for automatic detection of various plants diseases by utilizing a cascaded unsupervised image segmentation technique. In contrast to additive RGB color model for digital imagery, we exploit CIELab colour model, which as a pre-processing step enhances each channel using proposed next peak contrast stretching method. We also propose multilevel segmentation method where Initial segmentation is performed using expectation maximization algorithm with the constraint of minimum visual information loss. Ultimately, salient region is extracted out of the quantized image by employing binary partitioned tree which utilize largest principle eigen vector. Post-processing methods are performed for the elimination of inutile fragments, and resolutely image labeling disentangles the vital salient focus. This procedure is predominantly valuable when salient region in the image belongs to a single class, even with large distribution of intensities. The experimental analysis indicates that the new cascaded design results in superior color segmentation with the affirmation of infected regions extraction.