旨在探讨基于车载相机和HLS时序遥感数据相结合的作物分类方法，以提高作物分类的效率和精度，解决传统方法中地面样本采集效率低、遥感物候特征利用不充分等问题。研究以江苏省秋粮分类为例，验证了此方法的可行性和应用潜力。通过车载相机采集道路两侧作物图像并构建大量人工标注样本，结合2023年和2024年的HLS时序数据，采用高斯滤波重构时间连续的地表反射率，提取多维特征，构建随机森林分类模型。研究结果表明，水稻、玉米和大豆在HLS时序数据中表现出较为明显的差异。水稻的分类精度最高，生产者精度与用户精度均超过90%，而玉米和大豆因物候特征的相似性，精度相对较低（74%-85%）。模型在独立验证县的总体分类精度为89%，验证县内的水稻主要分布于全县东南区域，玉米和大豆则集中于西北区域，且分布特征清晰。车载相机结合HLS时序数据可实现高效的作物分类，随机森林模型对高维特征整合与抗样本不平衡性具有显著优势。尽管模型总体表现优良，但在破碎地块和高云量区域仍有改进空间。未来需融合多源遥感数据缓解云干扰，并扩展作物类型以增强模型泛化能力。

This study aims to develop a crop classification method by integrating vehicle images with HLS time-series remote sensing data. The goal is to enhance classification efficiency and accuracy, addressing the limitations of traditional methods such as low efficiency in ground sample collection and insufficient utilization of remote sensing phenological features. A vehicle-mounted camera system was deployed to collect manually annotated crop samples along road networks, combined with HLS time-series data from 2023 and 2024. Gaussian filtering was applied to reconstruct the time-series imagery, and the Random Forest classification method was employed to classify three major crops: rice, maize, and soybean. Results demonstrated significant differences in the characteristics of rice, maize, and soybean in the HLS time-series data. Among these crops, rice achieved the highest classification accuracy, with both producer's and user's accuracy exceeding 90%, whereas maize and soybean had lower accuracies (74%-85%) due to their similar phenological characteristics. The overall classification accuracy in the validation area was 89%. The rice in the verification area is mainly distributed in the southeast region of the county, while corn and soybeans are concentrated in the northwest region, and their distribution characteristics are clear. The integration of vehicle images and HLS time-series data proves effective for crop classification, with the Random Forest model demonstrating superior performance in handling high-dimensional features and sample imbalance. However, challenges remain in fragmented farmland and cloud-covered areas. Future improvements should focus on incorporating multi-source data to address cloud contamination and mixed-pixel effects in fragmented areas, while expanding crop categories to enhance model generalizability for broader agricultural applications.