Diseased Surface Assessment of Maize Cercospora Leaf Spot Using Hybrid Gaussian Quantum-Behaved Particle Swarm and Recurrent Neural Network

Cercospora zeae-maydis (CERCZM) is a destructive fungus that is strengthened by hot tropical weather and high humidity such as in the Philippines, resulting to recurrent adverse impacts of having maize Cercospora leas spot disease, and quantification of leaf damage is essential for plant phenotyping in understanding pathogen interaction. Visual detection of this disease often results in subjective classification. To address this challenge, the integration of computer vision and computational intelligence is employed in detecting healthy and damaged corn leaves and predicting the surface damage percentage due to maize Cercospora leaf spot. Dataset with 583 images contains matured healthy and diseased corn leaves that were grown outdoor and individually captured by a digital camera. Graph-cut segmentation through lazysnapping segmented the vegetation pixels and CIELab thresholding segmented healthy and diseased regions. Spectro-textural-morphological leaf signatures were extracted and selected using combined neighborhood component analysis and ReliefF resulting in R, H, *a, Cb, Cr, entropy, and whole leaf area. MobileNetV2 exhibited the best performance in classifying maize leaf health status. Gaussian quantum-behaved particle swarm optimized recurrent neural network (GQPSO-RNN) bested other feature-based machine learning and deep transfer image networks in predicting maize Cercospora leaf spot surface damage percentage with R2 of 0.949, RMSE of 6.290, and inference time of 3 seconds. This developed seamless MobileNetV2-GQPSO-RNN model provides reliable disease detection and quantitative assessment on the maize leaf surface in on-field phenotyping.