Frogeye leaf spot (FLS) is common in soybean cultivation and severely affects yield. Management practices for this disease include breeding resistant cultivars and applying fungicides. However, researchers often rely on traditional methods, such as visual identification, which are subjective and prone to errors. Machine vision and machine learning offer an alternative for FLS detection. This study proposed a segmentation method to extract disease spots from soybean leaves, using the hypergreen algorithm optimized by particle swarm optimization (PSO). A dataset consisting of 313 soybean leaf images, annotated with disease spots, was created. The PSO algorithm trained the data using a two-stage search strategy, comprising a coarse search phase and a fine search phase. The objective was to reduce the area discrepancy between predicted and annotated images by optimizing the weight values of the color components in the hyper-green algorithm. After optimization, the disease-to-leaf area ratio accuracy enhanced from 68.45% to 98.87%. The segmented images were analyzed, with 51 features extracted to construct a classification model. Three machine learning methods—random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBoost)—were utilized to create classification models for FLS, with all models optimized by the Bayesian optimizer. The RF and XGBoost models achieved accuracies of 94.76% and 94.22%, respectively, while the SVM model achieved the highest accuracy of 98.07%. The results indicate that this method serves as a reliable tool for the automated identification of FLS, enhancing both the efficiency and precision of soybean disease detection.