<p>There are many positives to using social media like Facebook, Instagram, and Twitter, but there are also many drawbacks. Cyberbullying (CB) is a problem that has arisen on these networks. The effects of cyberbullying on victims are difficult to quantify because responses vary widely depending on the individual. It is extremely difficult to identify bullying content in cyber messaging due to the vague nature of these communications. Reports of studies that use textual posts to tackle this problem have surfaced. However, less focus has been placed on detecting cyberbullying based on images. In this research, we discuss the findings of a comprehensive investigation of the dense region objects of cyberbullying images. Finding a well-suited model for the detection and classification of Cyberbullying images is a challenging problem. The primary objective of this work is to create a lightweight deep-learning model to combat the problem of cyberbullying using images on social media. In this work, initially, we designed a deep learning-based system trained on the 2-Dimensional Convolutional Neural Network (CB-2DCNN) to identify instances of cyberbullying images. Thereafter, we proposed and built Cyberbullying model based on You Only Look Once (CB-YOLO) is a soft prediction residual network model to effectively identify and detect cyberbullying in social media images with high accuracy, precision, recall, and f-score.</p>