Recognizing　handwritten　digits　is　a　crucial　yet　challenging　task　that　requires　exceptional　reliability　and　robustness　due　to　the　uniformity　of　digit　shapes　and　the　diverse　range　of　writing　styles　found　across　different　regions.　In　this　paper,　we　present　an　enhanced　version　of　the　LeNet-5　architecture,　meticulously　designed　to　address　the　complexities　of　digit　recognition.　To　simulate　the　unpredictability　of　real-world　scenarios,　we　introduced　noise　and　applied　preprocessing　techniques　to　the　MNIST　database,　thereby　increasing　its　complexity.　Our　proposed　model　underwent　rigorous　training　on　this　augmented　dataset,　resulting　in　the　development　of　a　comprehensive　recognition　system　that　seamlessly　integrates　both　software　and　hardware　components.　The　empirical　results　are　compelling:　our　refined　LeNet-5　algorithm　achieved　an　impressive　accuracy　of　at　least　88.8%　on　the　training　set　and　a　remarkable　88.9%　on　the　test　set,　even　in　the　presence　of　noise.　These　results　not　only　surpass　the　performance　of　other　leading　models　but　also　highlight　the　exceptional　robustness　and　broad　applicability　of　our　methodology　in　the　field　of　handwritten　digit　recognition.　©　2025　SPIE.