Gastric cancer is the biggest cause of cancer deaths globally; hence early identification is essential for treatment. Endoscopic pictures of malignant lesions are difficult to identify even for expert gastroenterologists owing to small differences in lesional appearance. To solve this, we offer a cutting-edge computer-aided detection method using deep learning and image processing. We use comprehensive literature to improve picture quality and segmentation accuracy to identify early stomach cancer. The Kvasir-SEG dataset comprises high-quality labeled endoscopic pictures. We use state-of-the-art preprocessing methods such contrast enhancement (CLAHE), noise reduction (Gaussian & Median Filtering), edge enhancement (Unsharp Masking), and normalization. The aforementioned methods improve lesion feature visibility before segmentation. We compare the performance of DeepLabV3+, a powerful deep learning model, with U-Net++, Swin U-Net, and EffiSegNet for segmentation. To identify lesions accurately, we compare them using IoU (Intersection over Union), Dice Coefficient, Sensitivity, and Specificity. Our method helps clinicians discover and recognize stomach cancer lesions in endoscopic images, improving patient diagnosis and prognosis. More biopsy findings and clinical history are planned for an improved diagnosis system. This AI-assisted, accurate, and cost-effective stomach cancer diagnosis method might revolutionize detection.