Using Residual Network for Plant Leaf Disease Image Classification | ||||
| New Valley University Journal of Basic and Applied Sciences | ||||
| Volume 3, Issue 1, June 2025, Page 34-45 PDF (1.76 MB) | ||||
| Document Type: Original papers | ||||
| DOI: 10.21608/nujbas.2025.370104.1033 | ||||
 View on SCiNiTO | ||||
| Authors | ||||
Ayman M Mahmoud   ; Romany F Mansour; Gamal F Badran
| ||||
| Department of Mathematics and Computer, Faculty of Science, New Valley University, El-Kharga, 72511, Egypt | ||||
| Abstract | ||||
| Many people in charge of agriculture field especially orchards, use examination to classify the quality of their harvest visually with expert eyes. This process is often very expensive and requires an abundance of labor, inaccurate due to lack of experience in some cases as well, however, with modern technology we can use deep learning for image classification. Generally, the industry is not always interested in the most accurate technique for a given problem but, in many cases, it is interested in the most appropriate technique for the expected result from that given problem. There should also be a balance between the computational cost and the accuracy. This paper uses deep learning with residual neural network to classify some common fruit tree leaf diseases such as apple scab, apple black rot, peach bacterial spot, potato late blight, and grape black measles, … using their digital images. We have adopted the resNet18 for image classification with Cross Entropy Loss as a loss function. Our optimizer here is Stochastic Gradient Descent. We compare different batch sizes as instance 1, 8, 32, 64, …. The result of changing the batch sizes highlights the improvement of the model performance at the expense of time. We choose 64 as batch size then we compare different values of learning rates starting from 0.01 to 0.05. When we reach the value 0.04 for the learning rate our model achieves a good classification accuracy of 95.1%. Finally, pointing out our future vision to enhance our model to increase accuracy. | ||||
| Keywords | ||||
| Deep Learning; PlantVillage; Neural Network; Classification; Residual Network | ||||
|
Statistics Article View: 94 PDF Download: 71 |
||||
