DATA AGGREGATION BASED HYBRID DEEP LEARNING TECHNIQUE FOR IDENTIFYING THE UNCERTAINTIES AND ACCURATE OBJECT DETECTION
Sreehari Patibandla, Maruthavanan Archana, Rama Chaithanya Tanguturi · Indian Journal of Computer Science and Engineering · 2022
The deep learning concept for performing object detection plans to minimize the labeling cost by identifying the samples which increase the detection into the unlabeled pool.According to the object detection than the classification process as the designing and selection of procedure is very essential.The related works have been implemented the aggregation data process of several outputs and batch box process for improving the performance evaluation.The mean Average Precision 𝒎𝑨𝑷 is the main performance metric for identifying the accuracy of the object detection.The class imbalance problem has been solved using the background class in every group of sample images.The loss related weight algorithm for training group is proposed in this paper utilizing the batch boxes, aggregating data and also the 𝒎𝑨𝑷 enhancements are addressed to solve the class imbalance problem.Additionally, a sampling process is used for identifying the uncertainty and enhancing the object detection process.The performance results illustrate that the proposed framework generates good performance than the relevant technique and it will be used for realtime applications in an efficient manner.