This paper introduces the creation of a medical assistant system aimed at improving the efficiency and precision of lung field segmentation and pneumonia Recognition. The system leverages advanced deep learning techniques, employing the U-Net++ model for lung field segmentation and an improved Inception-v3 network for pneumonia recognition. The U-Net++ model, selected for its ability to capture multi-scale features through nested skip pathways, achieved a Dice coefficient of 0.967 and an IoU of 0.938, demonstrating superior performance in segmenting lung regions accurately. For pneumonia recognition, we propose an improved Inception-v3 network that enhances feature extraction by integrating residual connections to address the vanishing gradient problem. The network incorporates both spatial and channel attention mechanisms to focus on relevant features, and uses channel shuffle to facilitate effective information fusion across different feature maps. This approach resulted in a classification accuracy of 94.64% and a recall rate of 99.72% for pneumonia cases. This system, supported by a user-friendly interface built with Vue.js and a robust backend developed using Spring Boot and Flask, provides an effective tool for radiologists, enabling faster and more accurate diagnoses while addressing challenges such as limited medical resources and an aging population.