Plant Seedling Classification Using Preprocessed Deep CNN

In developing and developed countries, farmers are struggling to reduce costs and provide organic produce. Farming large areas of land requires equipment, workers, and other material that burden farmers with increased costs to compete in the local, regional, and global markets. With the advent of new technologies in the field of Artificial Intelligence, Internet of Things (IoT), cloud computing, and others, there is a glimpse of hope for inventing new techniques in farming that will eventually reduce the cost of farming large areas of land. In this paper, a method is proposed that can automatically classify plant seedlings with great accuracy thus making it possible for automatic farming processes. We propose a Deep CNN architecture for the automatic classification of plant seedlings using whole images and using segmented images as input. The test accuracies on a dataset of 4722 images of 12 different species outperform similar methods reported in previous studies. The experiments showed that the proposed method achieved an average test accuracy of 91.58% when whole images are used as input and an average accuracy of 95.02% when segmented images are used as input to the proposed Deep CNN architecture. The segmented images increased the accuracy by 3.44%.