Fast measurement of crude protein content in fish feed based on visible and near infrared spectroscopy

The crude protein(CP) content is an important index to evaluate the quality of fish feed. Near infrared spectroscopy (NIRS) is often applied when a rapid quantization of CP content in feed is required. 240 samples of 4 fish feed brands (Takara sakana-ii, Shangpin, WEIYE and Clever fish) were collected for calibration models by partial least squares (PLS) and artificial neural network (ANN). Firstly, PLS models were developed with the comparison of different preprocessing methods, then certain selected PCs by principal component analysis (PCA) were used as the inputs of back propagation neural networks (BPNN) model. The prediction results showed that BPNN model was better than PLS model. The correlation coefficients were 0.9985. The overall results indicated that visible and near infrared spectroscopy combined with BPNN was successfully applied for the determination of crude protein content of fish feed. This would be helpful for the authenticity detection of fish feed and keep a fair competitive market management.