Options
2023
Journal Article
Title
Germinative paddy seed identification using deep convolutional neural network
Abstract
Paddy has been the staple food of millions of people and due to its rising demand, it is imperative to increase paddy crop yield within the limited availability of land. Paddy yield depends on many factors and one of the significant factors is the quality of germinative seed. However, the traditional methods for selecting quality germinative seeds are expensive and time-consuming. Therefore, automatic detection of germinative seeds using non-destructive techniques is unavoidable. Few methods exist for the identification of germinative paddy seeds. However, these methods suffered from low-accuracy and precision rates because the methods only depends on hand-crafted and classical features. In this research, we propose a novel framework to identify germinative seeds applying deep convolutional neural network (CNN). We collected paddy seeds’ images using a smartphone in open environment. However, acquiring images in open environment can lead to illumination, orientation, and scale-related challenges, which might hinder model performance. Therefore, we addressed the illumination problem by converting the images into HSV format and applying the image normalization technique to handle scale and orientation challenges. We also address the overfitting problem by deriving an optimal set of parameters by tuning the hyperparameters of the model. The experimental results on a dataset consisting of three paddy varieties: BRRI 36, BRRI 49, and BRRI 52 demonstrated that our proposed framework achieved high accuracy (99.50%) in identifying germinative paddy seeds. We also compare the performance of our model applying different pre-trained transfer learning techniques (googlenet, alexnet, resnet50, and resnet101) and an existing traditional feature-based technique, RSGES. Our proposed model significantly outperforms the transfer learning techniques and RSGES for all the datasets in terms of all evaluation metrics. Hence, this model can be integrated in the industry and farmer level as a non-destructive method for identifying germinative paddy seeds.
Author(s)