The classification of soybean seed quality is a critical challenge in the agricultural industry that can impact product quality and decision-making processes. This study uses a Convolutional Neural Network (CNN) with an InceptionV3 architecture to classify soybean seeds, utilizing color and texture feature extraction. Texture features were extracted using the Gray-Level Co-occurrence Matrix (GLCM), while color features were extracted from the Hue Saturation Value (HSV) and Hue Saturation Intensity (HSI) color spaces. The CNN model classifies soybean seeds into five categories: broken soybeans, immature soybeans, intact soybeans, skin-damaged soybeans, and spotted soybeans, using InceptionV3 without the top layer and adding pooling, dense, and dropout layers. All convolutional layers were frozen, and the model was compiled using the Adamax optimizer with a learning rate of 0.001 and the categorical_crossentropy loss function. The study involved an initial training phase of 10 epochs, followed by fine-tuning for 5 epochs. The initial training results improved accuracy from 44.09% to 84.38%. Fine-tuning for 5 epochs achieved an accuracy of 71.78%. The model evaluation on test data after the initial training showed an accuracy of 72.38%, with the fine-tuned model successfully predicting 18 new images correctly. Confusion matrix analysis revealed the highest F1-score for the spotted soybeans category (0.81), while the immature soybeans category had the lowest F1-score (0.63).