<p>Hyperspectral technology serves as a crucial means for simultaneously detecting internal and external defects as well as biophysical and chemical properties in potatoes, representing a key pathway to achieving integrated detection. However, a dedicated hyperspectral image-based detection method for external defects in red-skinned potatoes is still lacking. To address this gap, this study proposes a hyperspectral image feature-based method for detecting external defects in red-skinned potatoes. First, raw hyperspectral images within the 400 ~ 1000&#xa0;nm range were acquired and preprocessed. Subsequently, image processing was applied to the preprocessed hyperspectral images. Two-stage principal component analysis (PCA) extracted nine feature wavelengths, and the Otsu algorithm generated masks to precisely segment regions of interest (ROI). Feature extraction and modeling proceeded along two parallel paths: the hyperspectral image path computed directional gradient histogram (HOG) texture features within the ROI and quantified texture pattern differences among various defects using Pearson correlation analysis; the hyperspectral data path manually extracted the average reflectance of the ROI from the preprocessed hyperspectral image. Finally, Support Vector Machine (SVM) and Random Forest (RF) algorithms were employed to construct and compare classification models. The results indicate that models based on hyperspectral image features significantly outperform those based on hyperspectral data features. Among these, the “SVM-image features” combination emerged as the optimal model, achieving an overall accuracy rate of 96.4%. This study confirms the feasibility of applying hyperspectral technology to external defect detection in red-skinned potatoes. The established framework provides an effective technical solution for future integration with internal quality inspection and the realization of intelligent single-station sorting.</p>