High-resolution RGB image dataset for wheat seed varietal identification and purity assessment

Achieving maximum wheat yield becomes challenging when no optimal seed purity varietal identification protocol for a particular region is available. Recognizing and identifying seed varieties is still performed manually through direct visual inspection, which is labor-intensive, time-consuming, and prone to errors. Therefore, many researchers have turned to computer vision, machine learning, and deep learning models. However, this requires a sizable dataset of a specific location, which is limited and unavailable. To address this issue, we presented a publicly available, high-resolution wheat seed image dataset in collaboration with the Wheat Biotechnology Lab, Center for Advanced Studies in Agriculture and Food Security, University of Agriculture, Faisalabad, Pakistan, to solve this problem. Three wheat varieties- Akbar-19, Dilkash-20, and Urooj-22- each with 125 pure seeds- were selected because these varieties collectively account for 60-70 % of the nation's wheat production. The images captured for each seed variety are high-resolution RGB images taken under controlled circumstances to ensure uniform lighting and angles. Further, this paper discusses the impact of seed purity identification on wheat yield and the optimal purity of variety seed rates concerning wheat productivity. This paper emphasizes the critical role of region-specific/local varietal datasets in bridging the global AI innovation gap and advancing sustainable agriculture in all regions. The dataset provided is a valuable tool for scholars who wish to closely examine Pakistani wheat variety data and collection techniques, which could open up new avenues for research. Additionally, it promotes cooperation and raises the legitimacy of the existing research data in the scientific community, allowing for broader utilization.