New insights on the potentiality of deep learning techniques in supporting the assessment of some biotic diseases on maize

The convolutional neural network (CNN) architecture with minimum neural layers supporting plant disease identification is a promising tool. Existing models for plant disease identification take more deployment time and computation resources than CNNs with minimum neural layers. Maize is a source of food and feed worldwide and it is pivotal to address the factors that affect its yield. Among these factors, diseases caused by fungal pathogens play a key role. This study explored Inception-V3 CNN model by minimizing neural layers to identify the major fungal diseases like corn blight, common rust, gray leaf spot and stalk rot. The proposed model exhibited the average predictive accuracy of 96%. F₁ score of 0.91, 0.98, 0.91, 1.00, and 1.00 was observed for Corn blight, common rust, gray leaf spot, stalk rot, and healthy plants respectively. The recall ratio of corn blight, common rust, gray leaf spot, stalk rot and healthy plants was 0.94, 0.97, 0.89, 1.00 and 1.00 respectively. The precision for corn blight was 0.88, common rust was 0.99, gray leaf spot was 0.94, stalk rot was 1.00 and healthy plants were 1.00 respectively. The research helps in developing light weighted models without compromising the accuracy and adds machine-based support to the assessment of diseases in the agriculture.