Capsule endoscopy is a technique to capture images of the inside of the gastrointestinal tract by swallowing a device measuring approximately 11 mm in diameter and 26 mm in length. Compared with conventional endoscopy, capsule endoscopy is less burdensome on patients while allowing observations of the small intestine. This non-invasive technique produces more than 50,000 images in a single examination. Because a physician must visually check each image, a diagnosis is time consuming and labor intensive. However, some images captured by capsule endoscopy are dark, making diagnosis difficult, while others are covered with intestinal fluid and are unsuited for diagnosis. This study investigated automatic extraction of non-diagnosable images and automatic detection of lesions to reduce the burden on physicians, preventing missed lesions and support diagnosis.Here, we propose a technique to extract four types of non-diagnosable images: low luminance images, poorly focused images, images with adhered intestinal fluid, and images of bubbles. In addition, we use YOLOv5 (You Only Look Once version 5), which is a general object detection model, to automatically detect polypoid lesions after training a model with lesion images. Specifically, we trained the model with 5,466 polyp images, including original capsule endoscopy images and ones obtained by applying image processing techniques to the original images. Then we evaluated the model’s performance with a discrimination experiment using a total of 191 images: 93 with and 98 without polyps. When the recall was 100% to ensure that no polyp lesion was missed, the accuracy, precision, and F-measure were 77%, 68%, and 81%, respectively. The processing time per image was 0.021 seconds, confirming the effectiveness of the system. In the future, we will train the model with additional images of other lesions and improve the precision rate.