Explainable Approach for Species Identification using LIME

Abstract

Plant identification has a wide array of applications in the fields of agronomy and the discovery of natural and medicinal products. This research aims to explore various deep learning techniques like InceptionV3, Xpection, and ResNet to identify plants. Highly accurate machine learning models generally lack explainability and interpretability. Neural networks are usually opaque systems and thus a direct understanding of the interpretations becomes necessary. We aim to remove this ambiguity of how the model reaches its conclusion by introducing Explainable AI (XAI) techniques. Explainability aims to break such barriers by diminishing the lack of transparency in Artificial Intelligence and Machine Learning models, thus taking a step toward making AI reliable. In this paper, Convolutional Neural Network has been used to identify Vietnamese medicinal plant images based on the characteristics of the leaves, stems and other parts of the plant. Upon identification, our paper also elaborates on how each model predicts which part of the image helps the CNN model to make a prediction by integrating Explainable AI (XAI) using the Lime package. Through this research, we generated images using LIME package which highlight pixels that determine the result of our plant identification process.