The global consumption of chicken meat has surged due to its affordability, versatility, and perceived health benefits, making quality and safety crucial for public health and consumer trust. This study developed a non-destructive, real-time method for classifying chicken meat quality by integrating portable digital microscopes with Convolutional Neural Networks (CNNs). High-resolution images were captured using a 1,000× WiFi-enabled digital microscope and analyzed with two advanced CNN architectures, AlexNet and GoogLeNet, to categorize chicken meat into four classes: fresh, carrion, rotten, and formalinized. The methodology included systematic sampling and image preprocessing techniques—such as histogram equalization, noise reduction, and color space transformation—to enhance image quality and model performance. A dataset of 2,000 images was split into training and validation sets, with 600 images reserved for testing. Models were optimized using various hyperparameters, including optimizers Stochastic Gradient Descent with Momentum (SGDM), Adaptive Moment Estimation (Adam), Root Mean Square Propagation (RMSProp), and learning rates (0.0001, 0.00005). Results showed that GoogLeNet, optimized with RMSProp and a 0.00005 learning rate, achieved the highest testing accuracy of 99.15%, outperforming AlexNet’s 98.65%. The study highlighted that adaptive optimizers and lower learning rates significantly improve model accuracy and stability. Confusion matrix analysis confirmed high precision in classifying most categories, with minor errors in the rotten category. This approach enhances food safety standards, reduces the distribution of low-quality meat, minimizes food waste, and improves supply chain traceability. The CNN-based system offers the poultry industry a rapid, accurate, and cost-effective solution for automating meat quality assessments, boosting consumer confidence, and supporting global sustainability goals.