The paper presents a dual phase method based on Convolutional Neural Networks (CNNs) for detecting plant diseases in small rice grain datasets with heterogeneous backgrounds. The proposed approach addresses the challenge of limited training samples and the wide variety of backgrounds. In the first phase, the Faster RCNN method is used to extract the significant portion of the image, specifically the rice grain, resulting in a secondary dataset that excludes the heterogeneous background. In the second phase, disease classification is performed on the simplified samples using CNN architectures. The effectiveness of the dual phase approach is demonstrated by comparing it with the straightforward application of CNN or Faster RCNN on the small grain dataset. The proposed method achieves a five-fold cross-validation accuracy of 88.11%.

The paper titled "Dual Phase Convolutional Neural Network Based System for Small Rice Grain Disease Identification" proposes a method for effectively identifying diseases in a small rice grain dataset with heterogeneous backgrounds. The authors address the challenge of using Convolutional Neural Networks (CNNs) for plant disease detection, which typically require a large number of training samples and struggle with diverse backgrounds.

The proposed method consists of two phases. In the first phase, the authors employ the Faster RCNN method to crop out the significant portion of the image, specifically the rice grain. This step helps in creating a secondary dataset that contains rice grains without the heterogeneous background. By removing the background, the dataset becomes more simplified and focused on the rice grains themselves.

In the second phase, disease classification is performed on the derived and simplified rice grain samples using CNN architectures. The CNN models are trained on this dataset to learn the patterns and features indicative of different diseases in rice grains.

To evaluate the effectiveness of the proposed method, the authors compare it with two other approaches: a straightforward application of CNN and Faster RCNN on the small grain dataset. The results show that the dual phase approach outperforms the other methods, achieving a five-fold cross-validation accuracy of 88.11%.

In summary, the paper presents a dual phase convolutional neural network-based system for disease identification in small rice grain datasets. By incorporating the Faster RCNN method for initial cropping and simplification of the dataset, the proposed approach demonstrates improved accuracy compared to direct application of CNN or Faster RCNN alone.

The paper introduces a novel approach for plant disease detection specifically designed for small rice grain datasets with heterogeneous backgrounds. By combining the strengths of Convolutional Neural Networks (CNNs) and the Faster RCNN method, the proposed dual phase method achieves enhanced accuracy. In the first phase, Faster RCNN is utilized to extract the significant portion of the image, namely the rice grain, effectively eliminating the heterogeneous background. This process results in a secondary dataset consisting of simplified and refined samples of rice grains. Subsequently, CNN architectures are employed to classify the rice grain samples for disease detection. Comparisons with traditional CNN and Faster RCNN approaches applied directly on the small grain dataset demonstrate the superior performance of the proposed method, achieving an impressive five-fold cross-validation accuracy of 88.11%. This research provides a valuable contribution to the field of plant disease detection, addressing the challenges associated with limited training samples and heterogeneous backgrounds in a unique and effective manner.