Refined remote sensing identification of bamboo forests in complex terrain areas can help understand the temporal distribution of bamboo forests and incorporate the ecological， economic， and social values of bamboo forests. In-depth analysis and the effective use of spectral differences and textural features in bright and shadow areas are key issues of the refined identification of bamboo forest information. In this study， we modified the “Film-Based & Class-Oriented” （FB-CO） algorithm and verified the effectiveness of the improvement using Sentinel-2A MSI images. In the “Modified Film-Based & Class-Oriented” （MFB-CO） bamboo forest information remote sensing extraction algorithm， the normalized shaded vegetation index （NSVI） is used instead of single-band thresholds to segment the forestland in bright and shadow areas， and a linear regression model is utilized to enhance the shadow area information. The BPNN， SVM， and RF classifiers are introduced to extract bamboo forests. The results show that the best segmentation thresholds for forestland in bright and shadow areas based on the NSVI and NIR are 0.41 and 0.23， with an Overall Accuracy （OA） of 96.00% and 83.50%， respectively. After the enhancement of shaded area information， the fitted model R² was greater than 0.82 for each band， the MRE was less than 5%， the mean value increased for all bands， and the standard deviation decreased. The OA of the bamboo forest extraction is 82.41% for the FB-CO algorithm and 86.51%， 88.43%， and 88.92% for the BPNN， SVM， and RF based on the MFB-CO algorithm， respectively. The latter values are better than those of the FB-CO algorithm. The results show that the MFB-CO algorithm effectively improves the extraction of bamboo forest information by enhancing the implementation of several key steps of the FB-CO algorithm， providing technical support for the refinement of bamboo forest identification.