Glycosuria level monitoring is a fundamental aspect of diabetes management, enabling proactive and personalized care to prevent complications, improve overall health, and enhance the quality of life for individuals with diabetes. Conventional glycosuria monitoring methods often involve invasive procedures, potentially risking the immune system, especially with repeated injections. The research focuses on a non-invasive method utilizing urine samples for glycosuria testing to address these concerns. The proposed method integrates image processing techniques, such as Gaussian filtering and image resizing, to optimize input data with machine learning approaches. Deep learning, especially convolutional neural networks is employed for their ability in feature extraction and pattern recognition. The system aims to achieve high accuracy in categorizing glycosuria levels into groups such as diabetes, prediabetes, and normal, providing a reliable and non-invasive alternative to traditional glycosuria monitoring techniques. This approach shows promise in enhancing patient compliance and overall health monitoring for individuals requiring regular glucose assessment. The accuracy of the proposed method based on CNN achieves 98.54% where the existing method based on support vector machine (SVM) achieves only 88.46%. it is evident from simulation results, the proposed method attained 11.39% improved accuracy when compared to existing SVM.