Improving the performance of a support vector machine by using features extracted from mobilenetv2 for identifying coffee plant diseases in pagar alam

Summary

This study applied deep learning feature extraction (MobileNetV2) combined with Support Vector Machine classification to identify coffee leaf diseases in Pagar Alam, Indonesia. The hybrid approach significantly outperformed conventional SVM, achieving 95% accuracy in distinguishing healthy leaves from leaf spot and rust diseases. The lightweight, real-time capable system is proposed as a practical tool to support early disease detection and improve local coffee productivity.

Regional applicability

This study was conducted in Pagar Alam City, Indonesia, where coffee production is economically important. The findings are directly applicable to Indonesian coffee-growing regions with similar agro-climatic conditions; transferability to United Kingdom settings would be limited, as UK coffee cultivation is negligible, though the machine learning methodology could be adapted for UK horticultural pest or disease detection systems.

Key measures

Classification accuracy, precision, recall, F1-score; comparison between MobileNetV2+SVM and conventional SVM on leaf disease identification

Outcomes reported

The study evaluated the accuracy of an SVM classifier enhanced with MobileNetV2 feature extraction for identifying coffee leaf diseases (healthy leaves, leaf spot, and leaf rust). The optimised model achieved 95% accuracy, 95% precision, 94% recall, and 94% F1-score on 180 test images, compared to 84% accuracy for conventional SVM without feature extraction.

Topic tags