Aiming at the problems that the existing monitoring methods are difficult to accurately monitor the individual water status of plants in a large area, and the canopy closure of kiwifruit orchard leads to lack of root domain soil water content (RSWC) monitoring methods. Multi-layer perceptron (MLP) and canopy vegetation index were used to predict RSWC at 40cm depth of kiwifruit Xuxiang during fruit expansion period (May-September). In the preprocessing of MLP training data, Pearson correlation coefficient was used as the correlation evaluation index between input (vegetation index) and output (RSWC), and one-way ANOVA was used as the significance evaluation index between input and output. Further considering the possible impact of canopy acquisition range on model accuracy, the data were divided into different scales for training and evaluation of MLP. The results showed that renormalized difference vegetation index (RDVI) and RSWC had the highest correlation and significance, the correlation coefficient and P value were 0.744 and 0.007, respectively. This index could be used as the input of RSWC inversion. The modeling data of RDVI at different scales showed that the model accuracy was strongly correlated with RDVI sampling area A and diagonal length L(R2 was 0.991 and 0.993, respectively). In order to maximize the model accuracy, the sampling area should be between 2.540m2 and 3.038m2. The MLP model established by using RDVI of this scale achieved the maximum accuracy (R2 was 0.638, RMSE was 0.016). The research result can provide a basis for the establishment of soil water content estimation method and orchard irrigation system design of non-contact kiwifruit orchard.