Hyperparameter optimization of deep CNN classifier for plant species identification using artificial bee colony algorithm

Tailoring a deep convolutional neural network (CNN) for an implementation is a tedious and time-consuming task especially in image identification. In this study, an optimization scheme based on artificial bee colony (ABC) algorithm so-called optimal deep CNN (ODC) classifier for hyperparameter optimization of deep CNN is proposed for plant species identification. It is implemented to a ready-made leaf dataset namely Folio containing #637 images with 32 different plant species. The images are undergone various image preprocessing such as scaling, segmentation and augmentation so as to improve the efficacy of the ODC classifier. Therefore, the dataset is augmented from #637 to #15,288 leaf images whose #12,103 images is allocated for training phase and the remainder for testing the ODC. Moreover, a validation process on 20% of the training dataset is performed along with the training phase in both optimization and classification stages. The accuracy and loss performance of the ODC are examined over the training and validation results. The achieved ODC is verified through the test phase as well as by a comparison with the results in the literature in terms of performance evaluation metrics such as accuracy, sensitivity, specificity and F1-score. In order to further corroborate the proposed scheme, it is even subjected to a benchmark with optimization-based studies such as genetic, particle swarm and firefly algorithms through MNIST digit-image dataset. The ODC identifies the leaf images and digit-images with the best accuracy of 98.99% and 99.21% surpassing the state of the arts. Therefore, the proposed ODC is effective and useful in achieving an optimal CNN thanks to ABC algorithm.