CNN-based medicinal plant identification and classification using optimized SVM

The exact and unfailing categorization of medicinal plants exceeds the capabilities of the average individual because it necessitates in-depth subject expertise and physical detection is cumbersome and imprecise owing to human mistakes. There have been multiple efforts to automate the recognition of medicinal plants using images of plant parts like flowers, leaves, and bark. The most trustworthy data source, according to research, is Leaf. An Enhanced Convolutional Neural Network architecture (using modified ResNet50) with Progressive Transfer Learning (ECNN-PTL) has been proposed in this paper. The suggested method uses an improved ReNet50 framework for feature extraction along with PTL. Classification has been done using an Optimized Support Vector Machine (OSVM). The classical SVM hyperparameters are tuned further by the Adam optimizer to achieve a better performance model. During the first stage of training, the initial levels of the pre-trained ResNet50 architecture have been frozen while the recently introduced levels have been taught using a differentiated learning rate. In the second step, the refined model from the first stage is loaded and trained by restructuring. This technique has been replicated so that in these two learning steps, the image size is allowed to gradually rise from 64, 128, and 150 to 256 pixels. The proposed ResNet-50 effectively max-pools the activation from the previous fully connected layer to the subsequent convolution layer. In the trials, the maximum and average activations from the previous convolution are kept, giving the model knowledge of both the approaches and enhancing performance. The Indian Medicinal Plants Database (IMPLAD) has been used to compile the list of online medicinal plant species.The improved ResNet50 modelOSVM classifier in the ECNN-PTLapproach has been compared with baseline models like VGG16, VGG19 and ResNet50 in terms of accuracy, precision, recall, error rate and execution time. The modified ResNet50 + OSVM model achieve a testing phase accuracy of 96.8% and a training phase accuracy of 98.5%.