BNPL-Dataset: A New Benchmark Dataset for Visual Disease Detection of Barberry, Jujube, and Pomegranate Trees

Abstract

Leaf diseases in agriculture can be challenging to detect in a timely manner due to factors such as lack of manpower, poor eyesight, and quarantine restrictions. To address this issue, convolutional neural networks (CNNs) are a promising solution. However, the performance of CNNs depends on large datasets, which are often scarce for local species.To address this problem, we introduce a new dataset, the "Birjand Native Plants Leaves (BNPL) Dataset," which contains images of healthy leaves and pests and diseases affecting three common plants in South Khorasan province: Barberry, Jujube, and Pomegranate. The dataset includes 9 classes, with a large volume of data, making it suitable for training CNNs. We conducted experiments with several popular CNN architectures and gradient descent optimizers on the BNPL dataset. The results showed that the architectures, along with the optimizers, exhibited acceptable performance in classifying leaf diseases. Also, the BNPL dataset is publicly available to researchers.