Smart Diagnosis of Malabar Spinach Diseases Using ML and Deeplearning

Abstract

Malabar spinach (Basella alba) is a healthy vegetable that is actively grown in Bangladesh and is a source of many vitamins, minerals and antioxidant. But the productivity of leaf is greatly hit by diseases that lower yield and economic returns of farmers. Quality crop production and sustainability requires early and correct disease detection. This study provides a deep learning-based leaf disease detection model of Malabar spinach with the following diseases; Anthracnose, Bacterial Spot, Downy Mildew, Healthy Leaf and Pest Damage. We gathered 4,221 pictures, 1,215 of which were field pictures and 3,006 online pictures. Background removal and cropping followed, then the dataset was curated and 2,531 usable images were partitioned into 70% training, 20% validation and 10% testing. We adopted ViT-B/16 as the classifier and YOLOv8n and YOLOv11n models in terms of classifying objects. Accuracy, precision, recall, F1 score, and mAP at 0.5 were used as a measure of performance. ViT model scored 79.53 percent accuracy, which was a highest result in the Healthy Leaf classification. YOLOv8n with a mAP at 0.5 of 0.914 and a peak F1 of 0.83 was the best at Pest Damage recall. YOLOv11n was better at precision with an mAP of 0.895 and F1 of 0.81 on Anthracnose. YOLOv8n is suitable in tasks based on recall, YOLOv11n in precision and ViT in Healthy Leaf verification. The framework allows detecting diseases in real time, which will contribute to sustainable agriculture by minimizing the use of pesticide.