Advancing Tassel Detection and Counting: Annotation and Algorithms

Tassel counts provide valuable information related to flowering and yield prediction in maize, but are expensive time-consuming acquire via traditional manual approaches. High-resolution RGB imagery acquired by unmanned aerial vehicles (UAVs), coupled with advanced machine learning approaches, including deep (DL), provides a new capability for monitoring flowering. In this article, three state-of-the-art DL techniques, CenterNet based on point annotation, task-aware spatial disentanglement (TSD), detecting objects recursive feature pyramids switchable atrous convolution (DetectoRS) bounding box modified improve their performance application evaluated tassel detection relative Tasselnetv2+. The dataset the experiments is comprised of images maize tassels from plant breeding experiments, which vary size, complexity, overlap. Results show that annotations more accurate simpler than boxes, box-based approaches sensitive size boxes background point-based Overall, has high accuracy comparison other DetectoRS can better detect early-stage tassels. results these were robust Tasselnetv2+, number image.