High tide causes rip waves that causing disruptions and deaths in beaches of India and rest. In most of the beach drowning deaths are rising due to a lack of early warning information. Currently beach guards need real-time beach monitoring tide warning systems to rescue drowning people. While deep learning technologies excel at predicting objects, they struggle to accurately forecast high and low tides information for beach swimmers. At present the high and low tide detection accuracy is lower, due to that the early warning system are not functioning effectively. To improve the tide detection efficiency the dataset training must achieve higher accuracy. This paper addresses deep learning training issues to improve novel tide dataset training accuracy with novel tide dataset. This study suggests the best deep learning training network for beach tide classification. The work fine-tunes optimizers and epochs to look at the modern deep learning algorithms ResNet-18 and ResNet-50. This study tests deep learning training networks namely, RMSProp, SGDM and ADAM with epochs starting from 30 to 500 and applies three optimizers to balanced tide data. When using SGDM at shorter epochs, ResNet-18 and ResNet-50 achieved 100% training accuracy. The ResNet-50 training network had 100% classification accuracy with all three optimizers in lower and upper epochs. ResNet-50 integrated with SGDM and ADAM optimizers obtained 100% success at reduced epochs compared with ResNet-18. The present study examines only two training classes, i.e., high and low tides, and it can be extended by adding a few more object classes like humans and ferries. This unique approach aids in automating smart beach monitoring devices, enabling them to continuously send out high and low tide alerts using ResNet-50. The dissemination of tide information is crucial for rescue operations to prevent drowning cases and reduce fatalities in Indian and rest beaches.