Digital technology is accelerating the deep transformation of agriculture, and improving production efficiency has become the key to the high-quality development of agriculture. Crop growth prediction, as an important link in precision agriculture, can effectively guide agricultural production decisions and improve yield and quality through the integration of intelligent algorithms. In this paper, we constructed a BNM-PNN model combining Bayesian network and process neural network, and established a crop growth and development prediction model by collecting and analyzing the hyperspectral data, SPAD data and leaf area index of Italian shoot-tolerant lettuce under different nitrogen fertilizer application conditions. The study adopts the improved neural network initialization method and learning algorithm to solve the output locking problem of the Sigmoid-type excitation function and improve the convergence speed of the model. The results show that the BNM-PNN model has superior performance in crop growth prediction, with the coefficient of determination and regression estimation errors reaching 0.927 and 1.436 g/pl, respectively, with an average relative error of 3.24%. In the model performance evaluation, the accuracy of the BNM-PNN model reaches 96.15%, and the DICE score and sensitivity are 92.15% and 91.78%, which are significantly better than the traditional deep learning models such as GoogleNet and ResNet-50. The study shows that the fusion of Bayesian network and process neural network can effectively extract the key features of crop growth and provide technical support for accurate management of agricultural production.
