Performance of Image Recognition with Machine Vision: A Systematic Review

Abstract

This study aimed to identify the existing techniques, applications, equipment, and technologies applied for recognizing images with artificial vision via a systematic review of the literature during the period 2020-2022. PRISMA was used for selecting and analyzing 142 articles obtained from the EBSCO, Engineering Source, ProQuest, and ScienceDirect databases. Studies that were not directly related to the proposed objectives were not included, leaving 28 articles for full-text review. The review results strongly suggest that Hopfield-type convolutional artificial neural networks are highly effective for image recognition and classification tasks. Similarly, the combination of technological tools such as YOLO, Roboflow, Python, and OpenCV shows that image processing and deep learning are driving new applications that improve the various performance metrics of these tasks. Therefore, artificial vision, unlike technologies that incorporate electronic devices with sensors, allows the interpretation of an environment with a high degree of representation of reality, confirming its robustness in the complexity of data processing.