Indoor plant classification using deep learning techniques

Abstract

The plants interior spaces are used to improve the quality of indoor air and looks; however, identification of specific plant species is a complex process. This paper discusses an attempt towards categorization of indoor plants using deep learning methodologies. To compile the dataset, 3000 images of 10 different plant species were taken from local nurseries and personal gardens. Every image was marked according to the plant species mentioned above, such as Heartleaf-philodendrons, Aglaonemas, Ball-Cactus, and others. Data augmentation strategies were employed in order to expand the number of data instances and improve model robustness. The dataset was also used for training and testing a variety of deep and transfer learning models that include VGG19, EfficientNetB4, EfficientNetB6, ResNet152 and custom CNN. The results of the experiments indicated that VGG19 provided the highest accuracy which was 99. 58% which proves that it is useful for indoor plant classification. The findings of this study suggest that deep learning strategies could be effectively applied to optimize indoor plant management practices and promote higher levels of environmental responsibility. It will be beneficial to conduct future studies to identify more features and circumstances to increase the accuracy of the classification models and use them in other fields