For fast in-situ assessment of tiller phenotypes in rice breeding, we introduce the TillerPET model, an improved transformer-based deep learning solution that permits phenotyping the number and compactness of rice tillers in images of post-harvest rice stubble. A rice tiller phenotype dataset covering three years of field data and four experimental sites across China was constructed to train and validate the model. TillerPET reports an R² of 0.941 for counting tiller number, demonstrating state-of-the-art performance on the proposed RTP dataset. Beyond its minimal errors in estimating tiller number, TillerPET also achieves an R² of 0.978 for characterizing tiller compactness. The two phenotypic parameters exhibit a high degree of consistency with expert breeders, offering reliable phenotypic indicators to guide further breeding.

A later heading date generally leads to higher grain yield in favorable ecological regions; however, grain yield reaches a limit as the heading date exceeds a certain threshold. Ghd7 is the first cloned major gene that regulates heading date, plant height and grain number. Here, we investigated the relationship between Ghd7 and florigen genes Hd3a and RFT1, to determine their roles in regulating heading date and grain number under different photoperiods. Our results revealed that under long-day (LD) conditions, Hd3a acts prior to RFT1 to promote heading while negatively regulating plant height and grain number. In contrast, Ghd7 positively regulates heading date, plant height, and grain number by inhibiting both Hd3a and RFT1. Under short-day (SD) conditions, the functions of Hd3a and RFT1 remain consistent with those under LD conditions, but Ghd7 does not inhibit their expression, resulting in a weaker phenotypic effect compared to Hd3a. Additionally, under both LD and SD conditions, increased Ghd7 expression enhances its inhibitory effect on Hd3a and RFT1, leading to later heading and increased grain number; however, once the heading date exceeds 94 d, grain number no longer increases. Moreover, the gn1a allele increased grain number by 16.5% to 42.5%, while combinations of the elite alleles from Ghd7, Hd3a, RFT1, and Gn1a significantly increased grain number by up to 240.9%. Therefore, we propose a new breeding strategy to optimize the heading date and grain number using the Ghd7Hd3aRFT1gn1a combination of Ghd7, Hd3a, RFT1, and Gn1a under LD conditions, and the Ghd7hd3aRFT1gn1a combination under SD conditions. This strategy improved the yield of the high-quality Northeast variety Kongyu 131 (KY131) by 69.1% in Beijing and 93.7% in Hainan. This strategy will greatly improve the efficiency of north-to-south adaptation in rice, providing theoretical guidance for expanding the geographical adaptability of rice varieties.