The invention provides a method for rapidly and non-destructively detecting the quality of beet seeds based on a near
infrared spectrum technology, and relates to the technical field of non-destructive detection of the quality of the seeds, the near
infrared spectrum of the beet seeds is collected, and the characteristic
wavelength is extracted through an improved competitive self-adaptive reweighted sampling
algorithm; meanwhile, depth features are extracted by using a customized
convolutional neural network, and the depth features and the depth features are fused to form a fusion
feature vector with high characterization force. Secondly, establishing a quality
scoring system based on key agronomic characters after single seed
sowing, and constructing a spectral feature and
quality score contrast
database; for to-be-tested seeds, the most reliable reference seed group is screened out from the
database by calculating the weighted similarity and adopting a dual dynamic threshold strategy. And finally, in combination with the
local space density of the reference seed and the correlation degree between the reference seed and the to-be-detected seed, carrying out credibility weighted
local regression calculation, accurately predicting the final
quality score of the to-be-detected seed, and realizing grading judgment.