Therapeutic foliage: A balanced dataset designed for automated identification of leaves of common medicinal plants

Leaf pattern recognition through automated methods has gained significant popularity in the field of computer vision in recent years as it plays crucial role in monitoring plant health. By effectively identifying leaf patterns, it becomes possible to prevent the unnecessary wastage of crop yield as leaves depict relevant information about plant health, making it a vital pursuit in agricultural research. Owing to the significance of this research area here, this study has designed a novel database namely, Therapeutic foliage by including healthy and diseased leaf samples belonging to four different types of commonly available plants namely, Ixora (West Indian Jasmine), Catharanthus roseus (Madagascar Periwinkle), Bougainvillea (Paper flower) and Epipremnum aureum (Devil's ivy) having immense medicinal properties. The newly developed database comprises of 1200 images of diseased and healthy leaves where 300 images of leaf samples are included from each type of plant. The images containing leaves included in this database possess complex backgrounds containing weeds and various other objects and are captured under varied illumination conditions using different camera angles. To the best of our knowledge, Therapeutic foliage is the first database which is designed by including leaves from these four species, none of the existing databases contain leaves of these plants. The effectiveness of this database is also validated in this work by performing classification using several deep neural networks like AlexNet, Vgg16, ResNet152, MobileNetv2, ShuffleNetv2, SqueezeNet, InceptionV3, InceptionResNetv2, DenseNet121, NasNetLarge, NasNetMobile, GhostNetv2, EfficientNetB7 and Xception. Out of all these networks, DenseNet121 has given the best results by registering an overall Test Accuracy of 99.5 % and giving impressing outputs for quantitative parameters like Recall, Precision and F1 score.