Tomato Disease Recognition Using a Compact Convolutional Neural Network
Emre Özbılge, Mehtap Köse Ulukök, Önsen Toygar, Ebru Ozbılge
Abstract
Detection of the diseases on tomatoes in advance and making early intervention and treating increases the production amount, efficiency and quality which will satisfy the consumer with a more affordable shelf price. In this way, the efforts of the farmers who are waiting for the harvest throughout the season will not be wasted. In this paper a compact convolutional neural network (CNN) is proposed for diseases identification task where the network is comprised of only 6 layers that is why it is computationally cheap in terms of parameters employed in the network. This network is trained by using PlantVillage’s tomato crops dataset which consists of 10 classes (9 diseases and 1 healthy). The proposed network is first compared with well-known pre-trained ImageNet deep networks using transfer learning approach. The results show that the proposed network performed better than pre-trained knowledge transferred deep network models and it is shown that there is no need to constitute very large, complicated network architectures to achieve a superior tomato diseases identification performance. Furthermore, to increase the performance of proposed network, data augmentation techniques are also employed during the network training. The proposed network achieves an accuracy, <i>F</i><sub>1</sub> score, Matthews correlation coefficient, true positive rate and true negative rate of 99.70%, 98.49%, 98.31%, 98.49% and 99.81%, respectively using 9,077 unseen test images. Our results are better than or similar to the results of the state-of-the-art deep neural network approaches that used PlantVillage database and the proposed method employs the cheapest architecture.