Feature Learning Based Campus Scene Recognition And Location

Computer vision is an interdisciplinary scientific field that involves how computers gain a high level of understanding from digital images.From an engineering perspective,it attempts to understand and program the tasks that the human visual system can accomplish.Computer vision responsibilities include approaches for acquiring,processing,analyzing,understanding digital images,and extracting high-dimensional data to produce decision information in the form of numbers.Scene recognition technology as a fundamental problem in the robot,computer vision applications,and learning features was used in this research to enable the system to automatically discover the representations required for feature detection and classification from the Dataset.This research studied essential Feature learning-based campus scene recognition and location belonging to the field of image localization.This research-based on the theory of preprocessing,feature extraction,classification,and evaluation to output the location of the image resulted from the scene recognition.The author first collected the Dataset followed by preprocessing,extraction of features.Then the combination of Scale Invariant Feature Transform and Histogram of Oriented Gradients features to achieve the classification images using Support Vector Machine in the scene recognition.The current research made use of the skimage library(scikit-image)for feature extraction.Equivalent to each key point in the image,there would be a SIFT descriptor with dimensions exact by the parameters step size(distance between descriptor sampling points)and radius(relating size of the scanning area).The author also extracts a standard HOG descriptor equivalent to the entire image at a different granularity(by making use of the parameters pixels?per?cell,cells?per?block,and orientations),effectually permitting us to select features at different scales.Experiments used of the standard SVM classifier with scene recognition.Current research uses the sckit-learn(sklearn)library for the SVM executions.The author did cross-validation by random,splitting the Dataset into a training set and evaluation set.Current research constructed preparing features vectors from the preparing split.Finally,the description of the usual accuracy,confusion-matrix as well as standard information-retrieval data such as Precision,Recall,and F-measure were carried out.The current experiment was developed by using python for easy writing and feature extraction and classification of location.Furthermore,the experimental results on learning features,scene recognition,and location techniques were obtained and interpreted.The current experiment achieved the final results resulted from a combination of features.The results proved prediction accuracy to be 77%.The average precision is 78%,average recall is 76%,and the average f1-score is 72% for ensemble learning features.