Artificial Intelligence Mobile App For Identification Of Cassava Diseases In Farm

The project aims to revolutionize disease monitoring through the use of artificial intelligence (AI), advanced sensor technology, and crowd-sourcing, connecting the global agricultural community to support smallholder farmers. Specifically, a Convolutional Neural Network (CNN) was employed in this study. CNN models offer promise in enhancing plant disease phenotyping, where traditional methods rely on visual diagnostics requiring specialized training. Deploying CNNs on mobile devices presents new challenges such as varying lighting conditions and orientations. Therefore, evaluating these models under real-world conditions is crucial for their reliable integration into computer vision tools for plant disease assessment.

Our approach involved training a CNN object detection model to identify foliar disease symptoms in cassava (Manihot esculenta Crantz). Subsequently, we implemented the model in a mobile application and assessed its performance using images and videos captured in an agricultural field in Nigeria, totaling 720 diseased leaf samples. We conducted tests for two severity levels of symptoms—mild and pronounced—within each disease category to evaluate the model’s effectiveness in early symptom detection.

Across both severity levels, we observed a decline in performance metrics, specifically the F-1 score, when analyzing real-world images and video data. Notably, the F-1 score decreased by 32% for pronounced symptoms in real-world images, primarily due to reduced model recall. Our findings underscore the importance of fine-tuning recall metrics to achieve desired performance levels in practical settings if mobile CNN models are to fulfill their potential. Additionally, the varying performance outcomes between image and video inputs highlight critical considerations for designing applications intended for real-world deployment