<p>Despite the remarkable performance of neural networks across various fields, their decision processes often lack transparency and interpretability. Consequently, explainable AI has become increasingly important. Many current studies on the interpretability of image recognition can analyze which regions within an image significantly influence the decision outcomes, but they struggle to simultaneously meet the requirements of interpretability and high accuracy. This paper introduces a neural network based on concept similarity in experience for decision, called the CProtoNet (Concept Prototype Network). Firstly, it incorporates a specialized network layer designed to learn the object features within the dataset. These object features are referred to as partial prototypes of the corresponding class for the image. Subsequently, a concept extractor within the network layer extracts the semantic concepts represented by these partial prototypes, termed prototype concepts. Finally, we impose specific constraints on the network to make decisions based on the similarity between prototype concepts. The experiments on the Stanford Dogs Dataset, the Category Flower Dataset and the CUB-200-2011 indicate that CProtoNet achieves a classification accuracy slightly higher than that of similar unexplainable networks. Additionally, CProtoNet can identify the concept prototypes that most significantly influence the neural network decisions. The experiments on the Stanford Dogs Dataset and the Category Flower Dataset.</p>