A CNN-based android application for plant leaf classification at remote locations

The Earth has witnessed the evolution of thousands of plant species in the kingdom named Plantae. Due to the diversity and subtle differences in each plant, it becomes difficult for a novice to identify a particular plant and to know the properties associated with it. We propose a classification model that can solve this issue by categorizing the input plant image. Our methodology can classify up to 79 different plant species found predominantly in Himachal Pradesh located in India. A Deep Learning-based model is used to carry out the classification. Our model is optimized to work efficiently without a live internet connection on smartphones and other devices with limited computational power. A total of 79 distinct classes were classified using the Convolution neural network DenseNet-161 model architecture with a testing accuracy of 97.3%. The application works on any android platform and can classify the input plant image with an average latency of 1.98 s. Our application built on this model assists farmers and locals to get in-depth knowledge about the species including the local name, scientific name, description, and the care requirements by uploading or taking a picture of the plant leaf.

Automated summary

This article introduces a classification model that can identify plant species, using deep learning to achieve a testing accuracy of 97.3% in categorizing 79 different plant species predominantly found in Himachal Pradesh, India. The application is compatible with the Android platform and is optimized to work efficiently without a live internet connection, allowing farmers and locals to upload or take pictures of plant leaves to obtain in-depth knowledge about the species.