Resnet Features and Optimization Enabled Deep Learning for Indoor Object Detection and Object Recognition

Abstract

Indoor object detection and recognition (IODR) helps to progress the quality life of the visually weakened person by helping them in indoor navigation along with their day-to-day activities. This research introduced the African Adam Optimization-based SqueezeNet (AAO-based SqueezeNet) for object detection and recognition. Moreover, Generative Adversarial Network (GAN) is exploited to detect the thing in the images with the assistance of a generator and discriminator in GAN. The features, such as texture and Resnet features are extracted in the detected objects. In addition, the objects are recognized using SqueezeNet based on the extracted features. Moreover, the training process of SqueezeNet is done using devised AAO algorithm, which is the concatenation of the African Vultures Optimization Algorithm (AVOA) and Adam optimizer (AO). Analysis is done based on the metrics, such as testing accuracy, precision, and recall using the Mendeley dataset and it accomplished the values of 0.905, 0.902, and 0.939, respectively.