Beach waste detection is a critical step toward addressing environmental pollution and preserving marine ecosystems. In this research, we explore the application of object detection models for the identification of beach waste in complex environments. We took two popular architectures, YOLOv8n and YOLOv11n, and fine-tuned them on a custom dataset. Later, we proposed a transfer learning-based modified YOLOv11n model by replacing the default SiLU activation function with the ReLU activation function in the convolution layers. Training the proposed model required significantly less time-3.207 hours compared to 3.753 hours for the original YOLOv11n model-representing a 14.55% improvement in training speed. The proposed model attained a mean average precision at 50% IoU (mAP50) of 0.831 and a precision of 0.846 on the test set. These findings emphasize the potential of ReLU activation function in YOLOv11n for faster training in object detection tasks.