| Performance of two deep and shallow convolutional neural network architectures for classifying of cucumber leaves diseases |
| Paper ID : 1285-NICAME1402 |
| Authors: |
|
Hossein Akhtari *1, Hossein Navid1, ali ghaffarnezhad2, Nayyer Etminanfar2 1Department of Biosystems Engineering, Faculty of Agriculture, University of Tabriz, Iran 2Department of Agricultural Machinery Engineering, University of Tabriz, Tabriz, Iran |
| Abstract: |
| Cucumber is one of the most consumed products. Performance and quality of the cucumber production is affected by various factors such as pests, insects, and various diseases. Diagnosing the diseases in early stages can reduce economic losses, and increase the quality of production. Agriculture is an important area for the implementation of common techniques based on machine vision. Deep learning is one of the different types of common techniques in machine vision, which has made significant contributions to the classification and identification of operations used in precision agriculture. In this research, convolutional neural networks (CNN) based on deep learning were used to identify and classify healthy and unhealthy cucumber leaves. ReseNet-101 and MobileNet-v3 architectures were used to train healthy and unhealthy leaves of cucumber. Evaluation data in two types were prepared and obtained from Tabriz University greenhouse. Despite being shallow and with a small number of training parameters, MobileNet-v3 architecture provided significant results. The accuracy of the presented architecture identification and classification was equal to 98/64. The use of this type of architecture will be very suitable for use in smartphones and embedded systems due to its light and shallow structure. |
| Keywords: |
| Precision Agriculture, Deep learning, Convolutional Neural Network, Plant disease recognition |
| Status : Paper Accepted (Oral Presentation) |
