Method and device for non-destructive sorting of fertilization information of hatching eggs before incubation

A method for non-destructive sorting of fertilization information of hatching eggs before incubation, comprising the following steps:

• S1:
• obtaining a first time sequence feature of hatching eggs and collecting spectrum data of the hatching eggs according to the first time sequence feature;
• S2:
• preprocessing the spectrum data in the step S1;
• wherein the step S2 comprises the following substeps:
• S21:
• performing standard normal variable transformation processing on the spectrum data;
• and
• S22:
• processing the spectrum data after the standard normal variable transformation processing obtained in the step S21 by using a linear filtering algorithm;
• wherein the step S22 comprises the following substeps:
• S221;
• determining a size of a filtering window using a linear filtering algorithm;
• and
• S222:
• shifting the spectrum data after the standard normal variable transformation processing obtained in the step S21 by using the filtering window in the smoothing substep S221, fitting the spectrum data in the window, and outputting the fitted data, which is represented as:
• x
• ¯
• k
• =
• 1
• 2
• ⁢
• w
• +
• 1
• ⁢
• ∑
• i
• =
• -
• w
• +
• w
• x
• snv
• (
• k
• +
• i
• )
• wherein xk is a value after the spectrum data are fitted, xsnv(k+i) is data after the spectrum data are subjected to standard normal variable transformation, k is a spectrum band position, and 2w+1 is a window size;
• S3:
• selecting the spectrum data preprocessed in the step S2 to obtain a feature band;
• S4:
• constructing a hatching egg classification neural network model and training the hatching egg classification neural network model;
• wherein the hatching egg classification neural network model comprises one position encoding layer, three encoding layers, one linear layer and one logic classification layer, the position encoding layer performs position encoding on the feature band according to a band position of the feature band in an original spectrum;
• the three encoding layers are configured to perform dimension increase and dimension reduction on an input feature band with position information, and perform information fusion on the input feature band with position information, the linear layer is configured to perform linear change on the feature band after information fusion output by the three encoding layers, and output a feature band with a specified dimension;
• and the logic classification layer is configured to classify the feature band with the specified dimension output by the linear layer;
• S5:
• classifying the hatching egg information according to the feature band in the step S3 by using the hatching egg classification neural network model trained in the step S4 to obtain fertilization information;
• and
• S6:
• obtaining a second time sequence feature of the hatching eggs, and sorting the hatching eggs according to the second time sequence feature, the first time sequence feature in the step S1 and the fertilization information obtained in the step S5.