Optimization of Quantum Dilated Convolutional Neural Networks: Image Recognition With Quantum Computing

Abstract

As computer vision tasks increasingly rely on Convolutional Neural Networks (CNNs) with ever-expanding parameter counts, the need for computational resources for model training is growing unsustainable, surpassing traditional computing hardware's progress. To address this challenge, emerging paradigms such as quantum computing are gaining attention as prospective alternatives for the future. This manuscript proposes Quantum Dilated Convolutional Neural Networks Revolutionizing Image Recognition with Quantum Computing (QDCNN-IR-QC). The first step is to use the MNIST dataset for the input pictures. Subsequently, Improved Bilateral Texture Filtering (IBTF) is used to preprocess the input pictures. Subsequently, E-LBP is used to extract pertinent features from the preprocessed pictures. In most cases, E-LBP does not show that optimization methods for picture recognition have been adjusted. Therefore, in order to adjust the E-LBP weight parameter, this paper suggests an ISMO optimization approach. Lastly, a new quantum architecture for picture identification is developed using QDCNN. To implement the proposed approach, Python is used. This is where metrics like F-Measure, accuracy, sensitivity, specificity, and precision are assessed. When compared to current techniques such as QOCNN-IR-QC, ANN-IR-QC, and QKNN-IR-QC, the proposed approaches provide 5.27%, 7.21%, and 8.23% greater accuracy, respectively, in terms of efficiency.