<p>Grading apples is an important aspect of postharvest management; it not only has a direct effect on the market value of apples, but it also has a critical impact on consumer satisfaction. Grading apples using deep learning machine learning-based systems is currently disadvantaged by the shortage of data, environment sensitivity to features, and the lack of capability to extract features outside the machine learning capabilities. The aim of this study is to aim and develop a two-part framework for a deep learning-based automatic apple grading model that combines an optimized version of Fuzzy C-Means (FCM) clustering with a Cat Customized Golden Eagle Optimizer (CC-GEO) for reliable and robust feature selection. The attributes that were selected included color, texture, and edge information that were extracted and combined and processed through a hybrid Transformer-Recurrent Neural Network (RNN) model architecture for grading. The hybridization of the model greatly enhanced both learning efficiency and grading classification stability. The experimental comparisons demonstrated that the proposed model executed high-grade performance and stability as compared to conventional CNN, the RNN, and Bi-LSTM models. The framework is an effective, automated, scalable, reliable, and accurate model for grading apple quality and offers improvement in the performance efficiency of the agricultural supply chain.</p>