Early and accurate identification of plant leaf disease is essential to reducing crop loss and improving yield. This paper presents an image-based leaf disease diagnosis framework that integrates structured pre-processing, hybrid Otsu–K-means lesion segmentation, colour co-occurrence feature preparation and convolutional classification, and that returns a treatment recommendation alongside the predicted disease class. The input image is first converted to grey scale, filtered to suppress noise and resized to a standard dimension. K-means clustering in the Lab colour space, thresholded by the Otsu criterion, then isolates symptomatic regions from the background and from healthy tissue, so that learning is concentrated on the informative part of the leaf. Discriminative colour, texture and shape descriptors are prepared from the segmented output and indexed for efficient retrieval, and two convolutional backbones — AlexNet and GoogLeNet — are trained for multi-class classification over five tomato leaf categories. Evaluation is carried out at two levels: an architecture-level accuracy comparison, and a class-level confusion-matrix diagnostic. At the architecture level, GoogLeNet attains 99.10% accuracy against 98.17% for AlexNet and 98.00% for the existing convolutional baseline, confirming the benefit of multi-scale inception features at one-fifteenth of the parameter count. The class-level diagnostic on a small held-out subset shows perfect separation of bacterial spot and mosaic virus but systematic confusion among early blight, healthy and late blight, which localises the residual error to the three visually similar classes rather than distributing it uniformly. The framework and the diagnostic together indicate where further work on segmentation quality and class balance would yield the largest return.

Atul Raj Dhairya, Prof. Nagendra Patel, "Design and Performance Evaluation of a Hybrid Otsu–K-Means and CNN Model for Multi-Class Leaf Disease Classification" International Journal of Scientific Research and Engineering Development, V9(5): Page(314-320) September - October 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.