In remote sensingimage analysis, the normalization process of multispectral index features is crucial to improve the models classification accuracy and generalization ability. This paper was based on the Google Earth Engine (GEE) platform, Sentinel-1 SAR radar remote sensing images and Sentinel-2 A optical remote sensing images were used as the data sources, and different classification scenarios were constructed by calculating the multispectral exponential features before and after the normalization of the texture features, topographic features, polarization features, Sigmoid function, and employing four machine learning classifiers, namely, Random Forest, Gradient Boosting Tree, Support Vector Machine and Simple Bayes which were used to conduct classification experiments to analyze whether normalization of spectral index features was beneficial to the recognition of Camellia oleifera forests. Subsequently, the constructed convolutional neural network (CNN) and deep learning model combined with the Bi-LSTM module were compared with the machine learning classifiers to analyze the effect of different models on Camellia oleifera forest recognition. Based on the sample points of five land types in Jian City in 2021, the classification scenarios suitable for Camellia oleifera forest recognition were applied to the sample migration in different years (2019, 2020, 2022, 2023) to analyze the incremental and spatial distribution of Camellia oleifera forest area in each year. The results show that the normalized spectral index feature combined with random forest achieve the highest recognition accuracy in identifying Camellia oleifera forests, with an overall accuracy (OA) of 99.02%, a Kappa coefficient of 0.983 7, a user accuracy (UA) of 95.31% for Camellia oleifera forests, and a producer accuracy (PA) of 93.74% for Camellia oleifera forests; The deep learning model of CNN series in the study has slightly lower accuracy than random forest classifier for Camellia oleifera forests recognition, in which the overall accuracy (OA) of the deep learning model combined with the Bi-LSTM module is 98.69%, the Kappa coefficient is 0.971 3 The user accuracy (UA) of the Camellia oleifera forests is 94.96%, and the producer accuracy (PA) of the Camellia oleifera forests is 93.17%; 2021 The planted area of Camellia oleifera forests in Jian reached 1 844 881 000 mu, of which Suichuan County accounted for 27.67%, the largest county in area; the distribution of Camellia oleifera forests planting decreases from high terrain to low terrain, and the planting sites are mostly located in the hillside land and self-retained land near the family farms and the planted area of Camellia oleifera forests is increasing year by year. The extraction method of Camellia oleifera forests proposed in this study can help realize the dynamic monitoring and management of Camellia oleifera forests, and the proposed sample migration method can effectively reduce the cost of sample collection and labeling.