Fusarium head blight (FHB), caused mainly by the Fusarium graminearum species complex, is a devastating cereal disease associated with yield losses and mycotoxin contamination. Early infection is closely linked to spore germination and germ tube elongation, yet conventional monitoring methods are labor-intensive and poorly suited for dynamic phenotypic quantification. We present EffiFormer-CGS, a three-module deep learning framework integrating object detection, key point localization, and phenotypic quantification for microscopic images of FHB spores. A dataset of 2381 images was generated from systematic experiments with triazole fungicides (Prochloraz, Prothioconazole, and Tebuconazole) across multiple concentrations and time points. Spores were annotated with bounding boxes and fine-grained geometric key points, enabling calculation of germination degree as the ratio of germ tube length to body length. EffiFormer-CGS achieved 90.8% mAP@0.5:0.95 in object detection and 91.4% mAP@0.5 in key point localization. All fungicides significantly inhibited germination, with Prochloraz showing the strongest effect. Predictions closely matched manual counts, with germination rate errors ≤5.18%. EffiFormer-CGS provides an efficient, automated, and high-precision approach for spore germination analysis, supporting high-throughput fungicide screening, resistance monitoring, and sustainable FHB management.
Wang et al. (Sun,) studied this question.