Indoor Greenery Identification: A ResNet-50 and SVM-Based Approach for Plant Species Recognition

The accurate identifications of houseplant species are very important applications in horticulture, biodiversity conservation, and plant disease management. This research paper presents an approach for classification of house-plant species by focusing on the identification of various plants commonly found in indoor environments. In the model, a hybrid architecture is used in which ResNet-50 is combined with an SVM classifier implemented using the PyTorch deep learning (DL) framework. It utilizes the Adam optimizer to train efficiently. The dataset consists of images of various house plants, categorized into 47 classes with a total of 15,131 images. The ResNet-50 DL architecture has been used and, as an SVM classifier, has performed multi-class classification. The dataset has been divided into training (80%) and testing (20%) sets. The model has been trained for 10 epochs. The proposed model has achieved an impressing accuracy of 98.80%, thereby proving to identify plant species in a more precise manner. As shown by the loss curve, training and validation loss decreased steadily throughout the process. This apparently indicates that the model converged appropriately. The results demonstrate how the SVM-ResNet-50 hybrid is capable of alleviating complex image classification tasks. The major challenge for this experiment was the model’s insensitivity to changes caused in images by lighting and angles. Further work in the future will focus on enriching the model through more augmentation techniques and data augmentation to enhance generalization and robustness. Another possible improvement of the model is to implement attention mechanisms to better concentrate on important features.