Advancing automatic plant classification system in Saudi Arabia: introducing a novel dataset and ensemble deep learning approach

Automated plant detection plays a pivotal role in various domains, including agriculture, environmental monitoring, and biodiversity conservation. In this paper presents a novel deep learning model specifically designed for classifying the diverse flora of Saudi Arabia. To accomplish this task, a novel dataset was created, named SaudiArabiaFlora Dataset, comprising samples from ten distinct types of plants found across various regions of Saudi Arabia. Our novel database provides an extensive range of plant species. The proposed model, named MIV-PlantNet, leverages the strengths of three well-established architectures: MobileNet, Inception, and VGG. By combining their unique characteristics, the model aims to achieve superior performance in terms of classification accuracy, precision, and F1-score. Extensive experiments were conducted to evaluate the model’s efficacy, and comparisons were made with state-of-the-art models such as MobileNet, Inception, and VGG. The results demonstrate that the MIV-PlantNet deep learning model achieved an outstanding accuracy of 99%. Moreover, it demonstrates remarkable precision at 96% and an outstanding F1-score of 98%, underscoring its robustness and reliability. To gain insights into the model decision-making process, we utilized visual explainable AI approaches, specifically SHAP (SHapley Additive exPlanations). This analysis reveals the essential elements contributing to model predictions, enhancing our understanding of the classification process and model behavior. The findings of this study have substantial implications, accurate plant classification in Saudi Arabia has significant implications for biodiversity preservation and ecological studies. Our Dataset and MIV-PlantNet model offers exceptional resources and valuable insights for automated plant detection in various fields.