Few-shot learning convolutional neural network for primitive indian paddy grain identification using 2D-DWT injection and grey wolf optimizer algorithm
2024
Nabin Kumar Naik | Prabira Kumar Sethy | A. Geetha Devi | Santi Kumari Behera
Rice (Oryza sativa) is extensively taken and sold throughout the world. The most common primitive paddy are Goswami, Jamuna, Jamut, Mamata, Pooja, Rambanbas, Sanghmitra, Sanjeevani, Sadhana, and Sannasriya. The primary objective of this study is to enhance the classification accuracy of few-shot learning datasets via the development of a reliable feature extraction approach and the Optimization of convolutional neural network (CNN) hyperparameters. Here, the 2D-DWT features were injected before the fully connected layer and added to the deep features of the CNN. Furthermore, the hyperparameters of the CNN layer were optimized using the Grey Wolf Optimizer (GWO). This optimization strategy enhances the recognition rate on a limited training dataset, and the injection of 2D-DWT increases the dimension of the feature vector, which further elevates the performance of the model. The performance of proposed model is examined with our own dataset, which is shared in mendely (sethy, PRABIRA (2023), “PRIMITIVE PADDY VARITIES OF ODISHA, INDIA”, Mendeley Data, V1, https://doi.org/10.17632/67zv95bzt8.1) and achieved accuracy of 98.50 %, sensitivity of 98.50 %, specificity of 99.83 %, precosion of 98.56 %, FPR of 0.17 %, F1score of 98.50 %, MCC of 98.35 % and kappa of 91.67 %.
显示更多 [+] 显示较少 [-]