TinyML Implementation on Microcontrollers: The Case of Maize Leaf Disease Identification

Maize, a crucial global food crop, faces various diseases affecting agricultural yield and quality, which must be monitored. This study employs advanced Convolutional Neural Network (CNN) architecture for efficient maize leaf disease detection and classification. The proposed CNN maize disease detection (MDD) development involves optimization, data augmentation, batch size, and fine-tuning strategies in addition to properly selecting and using maize leaf images from Kaggle, Harvard Dataverse, and Mendeley as Dataset. The detailed assessment of the CNN-MDD through splitting of data into training and testing sets reveals significant correlations among maize leaf classifications and the impact of data augmentation on images. After the training phase, the CNN-MDD was deployed on a Microcontroller Unit (MCU) simulation model. The deployed MCU model for the CNN-MDD optimization parameters recorded a latency of 7.60 ms, a Random-Access Memory (RAM) size of 726.60 KB, a classifier flash size of 344.70 KB, and an accuracy of 94.60%. These optimized parameters show a high compatibility rate with our MCU specifications. In addition, our proposed CNN-MDD MCU technology also emphasizes leveraging low-cost edge devices like Arduino Ble 33 Sense for practical crop pest and disease identification in sub-Saharan African countries like Ghana. This research was accompanied with deliverables including the coalition and creation of a Ghanaian maize crop pest and disease dataset and the development and deployment of an optimal and low-cost CNN-MDD MCU edge device. These efforts aim to provide local Ghanaian and Sub-Saharan African farmers with a swift, offline solution for MDD, fostering sustainable agricultural practices.