Img2Variety is an intelligent tool designed for the automatic identification of rice and maize varieties. It leverages high-resolution images of plants captured at various growth stages and from two viewpoints, integrating transfer learning and a mix-up augmentation strategy. This approach offers a non-destructive, efficient, and robust method for identifying plant varieties based on their unique visual characteristics.
For rice variety identification, sub-dataset of the dataset WGSR (containing 93 rice accessions, two viewpoints, and multiple growth stages), were used for model training and testing. The model achieved an optimal identification accuracy of 88.66% on the test set.
The training data and testing data can be downloaded here Download.
The best-performing identification model (GPU version) can be downloaded here Download.
More detailed information about this strain in i-traits
For maize variety identification, a processed and resized version of the dataset WGSM, containing 224 maize inbred lines, were used for model training and testing. The model achieved a best test accuracy of nearly 80%.
The training data and testing data can be downloaded here Download.
The best-performing identification model (GPU version) can be downloaded here Download.
