Effective fine-grained feature extraction and classification of solid materials using hybrid region convolutional neural networks

摘要

Difference between similar feature points is presented in the fine-grained classification, which depends on discriminative in extremely localized regions. Hence, the accurate localization of discriminative regions is the major challenge found in the fine-grained feature extraction and classification. The patch-based framework has been described to address this issue. The accurate patch localization is enhanced by the triplet of patches with the logical constraints, it minimnized the feature set. Therefore, the object bounding boxes are the only need for the proposed approach. This paper presents an effective fine-grained feature extraction and classification schemes for solid materials. The Fuzzy logic-Scale Invariant Feature Transform (FL-SIFT) is introduced for feature extraction. FL-SIFT based key points are taken for the classification is performed by hybrid Multilayer Perceptron with Faster Region Convolutional Neural Networks (MLP-Faster RCNN). A key advantage of fine-grained based MLP-Faster RCNN approach is, on average better in identification with FL-SIFT key points. The model is retrained to play out the recognition of four sorts of metal articles with the whole procedure taking 4 h time to clarify and prepare the new model per strong piece. The simulation is implemented on Python platform and the results are evaluated by several evaluation measures like specificity, accuracy, precision, f-measure, and recall. The performance outcomes are compared with the existing approaches and existing works. It shows that the proposed model achieved maximum outcomes than existing schemes in terms of accuracy 98.3%, Precision 96%, specificity 97.87% and it takes very low execution time 1.46 s.