Lightweight Convolutional Neural Networks for On-Device Classification of Rice Leaf Diseases

Rizky Pratama; Maria Simanjuntak

Abstract

Early detection of rice leaf diseases helps farmers reduce yield loss, but expert diagnosis is not always available in rural areas. This study compared lightweight convolutional neural networks for classifying four classes (bacterial leaf blight, brown spot, leaf blast and healthy) on smartphones. A dataset of 4,800 field images was split 70:15:15 and augmented. MobileNetV2, EfficientNet-B0 and a baseline custom CNN were trained with transfer learning and converted to TensorFlow Lite. MobileNetV2 achieved 94.6% test accuracy and a macro F1-score of 0.94, with a 9.1 MB model and 48 ms inference on a mid-range phone. EfficientNet-B0 was slightly more accurate (95.2%) but 1.8 times slower. MobileNetV2 offers the best trade-off for an offline mobile diagnosis app.

Keywords

deep learning
MobileNetV2
plant disease
rice
TensorFlow Lite

References

Ferentinos, K. P. (2018). Deep learning models for plant disease detection and diagnosis. Computers and Electronics in Agriculture, 145, 311–318.

Howard, A. G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., Andreetto, M., & Adam, H. (2017). MobileNets: Efficient convolutional neural networks for mobile vision applications. arXiv:1704.04861.

LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436–444.

Mohanty, S. P., Hughes, D. P., & Salathé, M. (2016). Using deep learning for image-based plant disease detection. Frontiers in Plant Science, 7, 1419.

Sandler, M., Howard, A., Zhu, M., Zhmoginov, A., & Chen, L.-C. (2018). MobileNetV2: Inverted residuals and linear bottlenecks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 4510–4520).