ABSTRACT
In this study, a novel two-stage approach to deep learning for the detection and identification of diseases in plants is been introduced . Its outstandingly good results underscore the potential impact this technique offers for the future of global agriculture. They used a dataset publicly available with more than 65,000 images of diseased and healthy plant leaves, and trained their model to achieve an accuracy rate approaching 99.71%. This extraordinarily high level of specificity in recognizing 20 different plant species and diagnosing 35 distinct diseases holds out promise to transform plant health maintenance. The import of this research goes further than the impressive numbers alone. In a world where crop diseases are a ubiquitous threat to the world's food supply, the dual-stage approach is scalable and practicable. Farmers everywhere will be able to avail themselves of its accuracy at the species level in diagnoses, using nothing more than the ordinary smartphone. This method can ultimately reduce crop losses, allocate resources more efficiently, and enhance agricultural sustainability. The present study represents an epoch-making contribution. Yet in evidencing such achievements it acknowledges that there are unavoidable limitations, for instance, potential biases in their data set or the problems posed by practical difficulties in real-life conditions. Nonetheless it provides a basis for future studies and improvements, lending hope to future generations of agricultural engineers that there is scope for innovation even in the most ancient field of human endeavor. Through this research, humankind has made another significant step toward greater stability and resilience in the world food system, where technology will come into its own as a human race striving against crop diseases.
