The effective identification of herbicide-resistant kochia in sugar beet fields is crucial for adopting sustainable weed management strategies. A study was conducted in a greenhouse and in field to record hyperspectral data of dicamba-resistant, glyphosate-resistant, and glyphosate-susceptible kochia biotypes in sugar beet. Hyperspectral data was captured within the wavelength of 400 – 1000 nm and preprocessed with Savitzky-Golay filter and Standard Normal Variate in a sequential order. Recursive feature elimination-random forest (RFE-RF) feature selection algorithm was used to select ten informative wavelengths bands from 224 bands. Subsequently, the selected features were trained on a fully connected neural network to classify dicamba-resistant, glyphosate-resistant, glyphosate-susceptible and sugar beet. The findings revealed that a combination of hyperspectral imaging and deep neural network can effectively distinguish sugar beet from herbicide-resistant kochia biotypes under varying environmental conditions. The trained deep neural network achieved classification accuracies of 93.27% in the greenhouse experiment and 98.76% in the field.