Deep learning and machine learning algorithms combined with hyperspectral imaging technology were used to establish a fast and nondestructive method for detecting umami intensity in grass carp to solve the problems of strong subjectivity，long time-consumption and sample destructiveness in the current detection methods of umami intensity.The competitive adaptive reweighted sampling method was used to select the feature wavelengths of spectrum after collecting the hyperspectral data in grass carp.The Gaussian-weighted multi-head attention network （GMANet） was developed.The model of detecting umami intensity in grass carp was established and optimized with the support vector machine regression （SVR），partial least squares regression （PLSR） random forest （RF），1D-ResNet and other algorithms.The results showed that the root mean square error of prediction （RMSEP） and the coefficient of determination of prediction （RP2） of GMANet network was 0.008 2 and 0.884 4，better than the RMSEP and RP2 of 0.007 7 and 0.818 8 in the optimal modeling method SVR in traditional algorithms.It is indicated that hyperspectral technology has a large application prospect in the detection of umami intensity and the GMANet network can make full use of the spatial image and spectral information of the samples.It will provide a new method for the further application of detection with hyperspectral image.