In response to the problems of insufficient data homogeneity and spatiotemporal dynamics in traditional ecological benefit assessment models for restoration, this paper designed a restoration ecological benefit assessment model based on multi-source remote sensing data. Firstly, the article designed the framework of the model, which was divided into two parts: data collection and data processing. The study used Principal Component Analysis (PCA) for data fusion, and preprocessed and extracted features from the data. By introducing long short-term memory (LSTM) networks, a spatiotemporal dynamic model was established to handle the temporal and spatial changes in the area to be repaired. The random forest algorithm was eventually utilized to enhance the prognostic efficacy of the model. The experiment showed that the model in this article performed well in testing, and the mean squared error (MSE) and mean absolute error (MAE) of the features after data fusion were both below 1; the trend and seasonal parts of the decomposed output results have a good fit to the original data, and the residual part has randomness. Compared with other algorithm models, the performance was the best, with an accuracy of 95.51% and an F1 value of 0.93. This model has a good effect on evaluating the ecological benefits of restoration, providing a reference for such research.