Akademik Veri Yönetim Sistemi
8TH INTERNATIONAL ISTANBULSCIENTIFIC RESEARCH CONGRESS, İstanbul, Türkiye, 12 - 13 Mart 2022, cilt.1, sa.3, ss.564-572
With growing concerns about the increasing population in the world and finite food resources, the application of the benefits of contemporary computing technology to improve the efficiency of agricultural fields is inevitable. CNN models today are outperforming traditional machine learning methods especially in image classification area. Consequently, CNN models are being used more and more prevalent day to day. In this study, in addition to contributing to agricultural activities, it is aimed to reveal the best method in the classification of plant seedling images by using prominent CNN models and to facilitate the work of farmers in the agricultural field. Using various CNN models, plant seedlings’ images that were gathered from online data science platform Kaggle classified. There are 12 different plant seedling classes in the data set and the total number of data samples are 4750. Data preprocessing and data augmentation both are applied to improve the performance of the models. Furthermore, learning rate reduced properly to fine tune the models. Root-mean-square error was used to evaluate the model performance. It is observed that augmentation and removing backgrounds of the images significantly improved the performance of the models. Accordingly, With the use of the InceptionResNETv2 model, the accuracy value was obtained as 96.55 percent. This result is a significant improvement comparing the previous works on the same dataset. The results are encouraging to say that this work will improve agricultural activities and help farmers classify plant seedlings.