Quantum Kernels for Difficult Visual Discrimination

Quantum machine learning has attracted much attention in recent years as one of the applications of quantum computing. In particular, classification using quantum kernels has attracted attention as a means of efficient classification by mapping to a feature space. We aim to use quantum machine learning to identify good and defective products in images including factory products, building, and farm products. Though a fruit apple looks delicious, it has a vine crack inside when it is split in two, it loses its commercial value. It is very difficult to distinguish normal apples from apples with internal vine cracks which are not visible on the exterior, using photographs of the exterior. In this study, an attempt was made to classify internal defects with classical and quantum kernels using binarized images of apples, which are difficult to distinguish with the naked eye. As a result, the accuracy of the classical kernel was less than 0.75. However, with the quantum kernel, an accuracy of over 0.94 could be obtained. The performance of the quantum kernel varied significantly depending on its type. We found that quantum kernel circuits having Hadamard, and control Ry gate have affected the construction of the learning model. We have demonstrated that high accuracy can be achieved by using quantum kernels for the classification of images that are difficult to discriminate visually.