Plants diseases diagnosis based on MobileNet and Inception Architectures

Food shortages are some of the direct consequences of plant diseases in Developing countries. Artificial intelligence combined with image processing can help respond to these threats. In this paper, Inception v3 and MobileNet v2 are presented for plant disease detection using smartphones. We implemented the proposed methods and conducted the detection experiments carried out on an open dataset of 54305 images of leaves containing 14 species, spread over 38 distinct classes of combinations [plants, diseases], including healthy plants. The models were also run on the improved version of this dataset including 87867 images, and where there was no imbalance between classes. The results show that the Inception v3 method outperforms MobileNet v2, a state-of-the art model in terms of precision both on the original dataset and its augmented version, which makes it a very useful advisory or early warning tool. It should be noted that the presented method could be further improved by integrating several other varieties of plant 1 species and diseases taken under real culture conditions from several geographical areas.