The world of underwater is one of the least explored and documented places on earth, making all the various domains in it untapped for us. This motivates many exploration teams and researchers to understand it. Most of the underwater research is done by autonomous machines. The images taken in underwater object detection are often of low quality because of the diffusion of light, presence of suspended particles, etc. making it difficult to detect and study the image. WaterNet is a deep neural network-based model that is used to enhance images before passing through Modified YOLOv5. This has sparked many innovative ideas over the years to enhance the images and make them easier to research. This paper presents an innovative approach to address this problem using YOLOv5. Building upon the architecture of the original YOLOv5 model, we introduce modifications to the Spatial Pyramid Pooling(SPP) layers by utilizing average pooling instead of max pooling. Using the proposed work, an increase in mAP of almost 4.89% was found compared to the original YOLOv5 model and an increase of 11% compared to the KPE-YOLOv5 model in the Trash-ICRA19 dataset. To evaluate the impact of this change, we made comparisons with the traditional model and proposed a modified model to check which performs better.