Abstract: Abstract: Because of the short growth cycle, large yield and high fertilizer requirements of facility crop, which were the characteristics causing high cost and high complement in cultivation medium, nitrogen (N), phosphorus (P) and Potassium (K) deficiencies frequent occurrence in the growth of facility crops. Accurate monitoring and diagnosis of nutrient content in facility crops during the growth process was very important -.In order to diagnose tomato nitrogen, phosphorus and potassium nutrition level more accurately, rapidly and stably, in the aspect of the spectral diagnostics technology， the changes of reflectance on characteristic wavelengths were taking into consideration to assess the nutritional status of crops. -. Sensitive bands were selected by using genetic algorithms. Then, the quantitative models of tomato nitrogen, phosphorus, potassium were established via stepwise regression, principal component regression and partial least squares method respectively based on reflectance spectra. The results verified that the stepwise regression models outperformed the principal component and partial smallest squares regression models of nitrogen and phosphorus, while principal component regression get the best models of phosphorus. The correlation coefficient R of the best models were nitrogen (0.9026) > phosphorus (0.8819) > potassium (0.8561) . The root mean square error (RMSE) were nitrogen (0.3191) < phosphorus (0.4978) < potassium (0.5128).Imaging technology can analyze the change of texture and other characteristics that were caused by plant nutrient deficiency. Texture features were extracted from images under - sensitive wavelength by using principal component analysis. The nutrients models of the tomato leaf nitrogen, phosphorus and potassium based on image features were established by stepwise regression, principal component regression and partial least squares method respectively. The results verified that the principal component regression models outperformed others models of nitrogen and phosphorus, while partial least squares method get the best models of phosphorus. The correlation coefficient R of the best models were nitrogen (0.9271) > potassium (0.8991) >phosphorus (0.8673). The root mean square error (RMSE) were nitrogen (0.3413) < phosphorus (0.3994) < potassium (0.5628).For overcoming the inadequacies of models build with single feature sauce, diagnosed models of multi-information fusion was established for tomato nutrients stress via artificial neural network modeling. Feature layer fusion was combining with the internal components and external morphological caused by crop nutrients stress. The correlation coefficient R of nitrogen, phosphorus and potassium were 0.9651, 0.9216 and 0.9353. The root mean square error (RMSE) were 0.19, 0.33 and 0.29. Results fully showed that the spectra reflection technology and image technology after feature layer integration models were better than spectra reflection or single image technology. Artificial neural network models of nitrogen, phosphorus and potassium improve the correlation coefficient R accurately were 6.25%, 3.97%, 7.92% than the single spectral model and 3.80%, 5.43%, 3.26% than the single image model. Furthermore, the detection root mean square error was reducing.The results showed that the multi-information fusion models achieve a substantial increase in model accuracy and have better diagnostic accuracy in achieving high accuracy compared with a single feature model, thus the rapid and high sensitivity detection of nutritional stress of the tomato leaves could be realized, which provides basis to methods about crop nutrients for the development of fast and accurate diagnostic instrument with important academic value and application prospect.