Dataset for Image-based Analysis of Mineral Fertilizer Granules

Dmitrii Iunovidov, Ikechi Kalu Ndukwe, Mohammad Reza Bahrami, Manuel Mazzara, Elizaveta Iunovidova

Keywords: Dataset, Industry, Fertilizer Granules, Quality Control, Instance Segmentation, Computer Vision

Abstract: In the context of the mineral fertilizer industry, a crucial sector for global food production, which faces challenges in production efficiency and fast quality control, this work introduces the Mineral Fertilizer Dataset (MFD), a novel annotated segmentation dataset comprising 1,608 images and 125,648 instances of various fertilizer granules with different colors. Addressing the lack of datasets in this field, the MFD supports both semantic and instance segmentation tasks, with segmentation masks that facilitate the computation of the equivalent area diameter of granules. Periodic checks of the area equivalent diameter based on customer specifications are essential to prevent potential defects, such as caking and dustiness, in the produced fertilizer granules. Baseline models based on Feature Pyramid Network (FPN), UNet, and MANet were trained for semantic segmentation, while baseline models based on Mask R-CNN, YOLOv8, YOLOv9, and Mask2Former were trained for instance segmentation. Our experiments demonstrate the efficacy of these models, as well as the robustness of the trained models in identifying fertilizer granules of different colors not included in our dataset, fertilizer granules under 365 nm ultraviolet light, as well as other granular objects such as Polyethylene Terephthalate (PET) pellets, corn, beans, and even pharmaceutical tablets. This dataset, along with its benchmark results on existing semantic and instance segmentation algorithms, aims to facilitate further advancements in computer vision applications for quality control in the fertilizer industry and related sectors.

Primary Area: datasets and benchmarks

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 10033