Construction Of Potato Moisture Diagnosis Model Based On Hyperspectrum

Potato is the advantage and characteristic industry of Inner Mongolia,but the lack of water and the lack of precise management technology seriously limited the development of the industry.Therefore,the research of potato water diagnosis technology based on hyperspectral was carried out.With Kexin No.1 as the experimental material,excessive irrigation(95% of maximum field water capacity),full irrigation(85% of maximum field water capacity)and moderate water shortage treatment(70% of maximum field water capacity)were set in tuber swelling stage from 2021 to 2022.Severe water scarcity(55%of maximum field capacity)and extreme water scarcity(45% of maximum field capacity).The indexes of yield,leaf area and biomass under different water treatments were tested,and the water-sensitive spectral bands were screened.A water diagnosis model based on the water content of leaves and aboveground parts was constructed.The main results are as follows:1.The yield,leaf area and biomass of potato tuber decreased gradually with the increase of water deficit during the tuber expansion stage.Both leaf water content and ground water content decreased with the increase of water deficit.After rewatering,leaf water content gradually recovered to the level of full irrigation,but ground water content was significantly lower than that of full irrigation.2.Through the correlation analysis with leaf water content,14 sensitive bands were selected;6 first-order differential variables;Seven spectral indices.14 sensitive bands were selected through the correlation analysis with above-ground water content.6 first-order differential variables;6 spectral indices were used to establish the characteristic spectral database.3.The partial least squares model,BP neural network model and support vector machine model are established by using the quantitative relationship between the characteristic spectral database of correlation screening and leaf water content and aboveground water content.The regression correlation coefficients of the six models are all greater than 0.5,and all of them can predict the water content of plants during tuber expansion.Among all the prediction models,the support vector machine model for predicting leaf water content has the best accuracy,the regression correlation coefficient is0.8526,and the mean square error is 12.2077,so it is the best water diagnosis model for predicting tuber water content during swelling period.