An Intelligent-Fusion Method of Product Core Feature Recognition Towards Design Patent Infringement Judgment

The identification of core features plays an important role of design patent infringement judgment (DPIJ). The traditional method usually depends on the experience and knowledge of on-site engineers and the will of judges. There are obvious uncertainty and subjectivity, which is prone to occur miscarriage of justice and then hinders technological innovation. For overcoming the problem, this paper first proposes a fusion method for intelligent recognition of product core features, which combines resnet50 model with Convolutional Block Attention Module (CBAM) attention mechanism and cosine similarity algorithm. First, design patents of target products are downloaded and then the images of multi-views are extracted to build a database. Following this, Canny operator is used to extract the contour features of images for further fileting data. Second, cbam-resnet50 model is trained and optimized by transfer learning approach to realize the quantitative recognition of target products’ salient feature regions. In addition, vgg16 model and cosine algorithm are integrated to calculate the similarity of contour features between target product and patent in database. Based on the two values, the target products’ core features are determined by max score algorithm. The bathroom shower is taken as example to verify the effectiveness of proposed method. The results show that cbam-resnet50 model is superior to other models in terms of Intersection over Union (IOU) and Dice Similarity Coefficient (DSC) indexes, and the vgg16 model plus cosine similarity method has the highest discrimination for the contour feature. Moreover, the accuracy of the proposed method is higher than traditional method and even LLMs.