An improved object detection and localization using deep convolutional neural network architecture

Abstract

The localization of objects is a significant computing technology in newlinewhich the goal is to evaluate accurate bounding boxes amongst objects newlinecontained in provided image. However, background clutter, occlusion, and newlineintra-class variations are some of the drawbacks which make object detection newlinea challenging issue. Object localization is applicable to numerous images newlineunderstanding tasks like object recognition, segmentation, and separating newlineforeground from background. Numerous object detection methods are devised newlinefor object localization, but it remains confined due to illumination changes newlineobject occlusion, perspective changes strong background noise, and cluttered newlineimages. With the aim of overcoming the limitations of the existing newlinealgorithms, two methods are devised for effective object localization. The two newlinecontributions of research are devised as: The preliminary contribution is to newlinedevise an effective object localization method, which is done using proposed newlineCat Crow Optimization (CCO). Here, the sparse-FCM is adapted for newlineperforming clustering with high dimensional data. In addition, deep newlineConvolutional Neural Network (deep CNN) is adapted to classify object using newlineproposed CCO algorithm, which is devised by combining Cat Swarm newlineOptimization (CSO) and Crow Search Algorithm (CSA). The proposed CCO newlinealgorithm determines tunes optimum weights in deep CNN for object newlineclassification. The second contribution is to perform object localization using newlineproposed Stochastic-Cat Crow optimization (Stochastic-CCO) based Deep newlineCNN. Here, the proposed Stochastic-CCO algorithm is used to train the Deep newlineCNN, which is devised by integrating Stochastic Gradient Descent (SGD) and newlineCat Crow Optimization (CCO) algorithm for attaining effective object newlineclassification. newline