Sporesight: A real-time computer vision-driven inference tool for pollen development stage classification in eggplant

Eggplant (Solanum melongena L.) is one of the major vegetable crops in the Philippines, significantly contributing to agricultural productivity and rural livelihoods. The economic contribution of this crop, valued at more than 9 billion pesos in 2024, underscores the need to continuously develop new and improved varieties that can adapt to the rapidly changing climate. One of the key strategies to expedite breeding activities leverages the use of doubled haploid technology, which requires the use of precise developmental stages of microspores or pollen for in vitro anther culture. This study presents Sporesight, a real-time, machine learning-driven desktop application designed to automate the classification of eggplant pollen developmental stages using object detection techniques. Initially, an expertly annotated dataset of 124 unique microscopic images, containing 3479 instances spanning seven distinct classes corresponding to eggplant microspore developmental stages, was used to train an AI model using the YOLOv5 algorithm. The model achieved a mean Average Precision (mAP@0.5) of 0.628, with high accuracy for morphologically distinct classes but moderate confusion for visually similar classes. This AI model was then integrated into an intuitive graphical user interface that provides image upload and preview, class-wise result visualization, and inference capabilities for the captured microscopic field of view, at an average time of 2.9 frames per second. As each captured microscopic field of view corresponded to a single frame, the system delivered inference results within 349 milliseconds. Sporesight provides high-throughput capabilities for selecting explants with suitable microspore developmental stages for in vitro culture, thereby contributing to streamlining the efforts to accelerate the development of climate-smart eggplant varieties.