Like many other plants, sugarcane is susceptible to diseases and if undetected and not addressed swiftly, can have deleterious effects on agricultural production and consequently, food security. It is imperative to promptly recognize leaf diseases and apply control measures to lessen the damaging effect on its agricultural yield. A base convolutional neural network was created to assess for sugarcane leaf disease with the following layout: input pixel = 152 × 114, epochs = 50, dense node = 64, drop-out rate = 0.3 and a learning rate of 0.0001 as parameters. Multiple experiments for various batch sizes were performed for the base neural network model and compared with pre-trained networks (VGG19 and InceptionV3). The best performing model generated adequate performance indices with F1-score (0.85–0.90), precision (0.87–0.89), recall (0.83–0.93) and accuracy rates (0.83–0.91) pointing to acceptable predictive capability to classify sugarcane leaf disease. The results were in conformity with the guidance of other reported research studies of utilizing smaller batch sizes as larger batch sizes did not yield superb performance as well as having the potential to give rise to immense computational expense. The adequacy of predictive capacity points to its feasible deployment in the actual agricultural workplace providing decision support to the agricultural stakeholders, thus leading to enhanced food safety and security. A functioning cooperation between agricultural stakeholders with data scientists is exemplary and advantageous for rapid and unambiguous diagnosis of sugarcane leaf disease leading to food security and safety.