Comparison of Conventional and Computer-Based Detection of Severity Scales of Stalk Rot Disease in Maize

Various diseases harm the maize crop, but stalk rot has significantly reduced crop yield.The susceptible stalk requires identification by pathologists to apply the precise dose of fungicide to the crop.Farmers in developing nations faced challenges for their timely hiring.Furthermore, differences in pathologists' professional competencies result in inaccurate diagnoses.In this paper, the convolutional neural network (CNN) utilization helped classify the severity levels of stalk rot as elaborated in Hooker's scale.The field experiment commenced at the Maize and Millet Research Institute Yousafwala, Sahiwal, using a smartphone to get images of resistant and susceptible lines fed to the proposed model for evaluation into six severity scales.The model's overall accuracy was 83.58%.Recording of the recall ratio of highly susceptible, susceptible, moderately susceptible, highly resistant, resistant, and moderately resistant had scores of 1.000, 0.766, 0.966, 0.800, 0.733, and 1.000, respectively, with an average of 0.877.Precision for highly resistant was 1.000, resistant was 0.785, moderately resistant was 0.789, moderately susceptible was 0.805, susceptible was 0.958, and highly susceptible was 1.000, with an average of 0.889.Highly significant (P < 0.01) results from the chi-square test exhibited significant differences between traditional and deep learning approaches.The results of the proposed model showed less confusion than the visual-based method.The proposed approach is a vital source of detection of resistant lines against stalk rot disease by developing country farmers.The suggested model eliminates the need for pathologists, making it a valuable tool for identifying stalk rot resistant lines.It aids farmers in finding resistant lines for breeding projects and estimating the fungicide dose against stalk rot.It also helps minimize the production cost and environmental pollution.