According to the
xerophthalmia core index multi-task AI detection
system, method and terminal provided by the invention, the
xerophthalmia multi-
modal image is collected and exclusive preprocessing is completed, and core index detection is carried out on the preprocessed
image based on the
xerophthalmia multi-index detection model, so that the xerophthalmia core index detection result is obtained. Outputting detection results such as
meibomian gland deletion rate, tear film rupture time,
staining classification, tear river height and
blepharitis marginal
inflammation classification; and dynamically adjusting the index result according to a
calibration coefficient corresponding to the type of the
data acquisition equipment, and finally generating a standardized detection report. A xerophthalmia multi-index detection model is constructed based on an improved U-Net3 + architecture, a dry eye feature enhancement
encoder and a cross-scale
feature fusion decoder are adopted, a customized multi-task detection head is matched, an inter-task attention
collaboration mechanism is established, meanwhile, a
preprocessing algorithm and a dynamic calibration mechanism matched with a dry eye image are designed, and the detection accuracy of the xerophthalmia is improved. And multi-index synchronous, high-precision and high-efficiency detection is effectively realized.