Classification of Poisonous Ornamental Plants Using CNN and ResNet Methods

Abstract

This research, titled "Classification of Poisonous Ornamental Plants Using CNN and ResNet Methods," defines poisonous ornamental plants as those containing toxic substances that can cause pain, allergies, or even death. These plants come in many different varieties, each with its own unique appeal. Many laypeople still find it difficult to distinguish poisonous from non-poisonous ornamental plants, especially in household environments that pose a risk to children and pets. Therefore, vigilance is needed in recognizing these plants. This study aims to develop a classification system for poisonous and non-poisonous ornamental plants using CNN methods with and without ResNet architecture. Furthermore, this study also aims to implement the trained model into a web application using Flask, so users can easily upload images of ornamental plants and obtain information about their potential toxicity in real-time. This research method uses deep learning techniques, specifically Convolutional Neural Network (CNN) with ResNet-50 and regular Convolutional Neural Network (CNN) or without ResNet with data divided into (70% training, 15% validation, and 15% testing). The test results for ResNet showed an accuracy of 98.25%, while the test results for regular CNN reached an accuracy of 87.47%. These accuracy results indicate that CNN with ResNet is superior for classifying poisonous and non-poisonous ornamental plants compared to CNN without ResNet