Automation through the introduction of robots is advancing in food processing plants, creating a need to grasp food items without dropping or damaging them. To achieve this, underactuated grippers that conform to the shape of the object are attracting attention. However, underactuated grippers face the challenge of difficulty in sensing the grasping state. While methods for sensing the grasping state exist that use internal sensors such as pressure sensors and bending sensors, these result in systems that are dependent on specific robot hands. Furthermore, in food processing plants, they pose a risk of foreign object contamination. Therefore, in this study, we propose a method that uses images captured by a monocular camera to sense the state of the underactuated hand and the object, thereby recognizing the grasping state without using internal sensors and enabling the hand to grasp the object without dropping or damaging it.