Smart irrigation systems are essential in advancing sustainable agriculture and mitigating global water scarcity in regions with limited water resources and unpredictable rainfall. Conventional irrigation practices, such as manual or time-based watering, frequently lead to sickly crops and water waste. This study recommends the implementation of a smart irrigation system that employs reinforcement learning (RL) to maintain soil hydration levels between 30% and 50% in order to optimise water efficiency. The system integrates real-time weather data from the OpenWeatherMap API, which includes temperature, humidity, and precipitation, using environmental parameters and evapotranspiration variables. Additionally, it simulates the dynamics of soil moisture. In contrast to conventional rule-based irrigation systems that are unable to adapt to changing environmental conditions, the proposed RL-based method reduces water wastage by 30% and maintains optimal soil moisture levels in 85% of cases. These enhancements are the result of a Deep Q-Network (DQN) algorithm that acquires knowledge about the surrounding environment and modifies its watering techniques accordingly. The results demonstrate that the model effectively reduces over and under-irrigation, thereby ensuring precise water management. With an average reward of 4996.8, this system is a scalable and dependable solution for sustainable agricultural practices, as evidenced by its consistent performance over numerous episodes and a low standard deviation of 6.4.