Abstract—Wireless communication is continuously evolving. Signals compete in the spectrum as there is a demand for data transfer. There is also a growing demand for radio spectrum, resulting in congestion. With the advancement of deep learning, there are numerous applications in which it could be utilized; this could be used as a solution to the current issue. Given the state of wireless communication, this research aims to achieve a higher accuracy rate regarding spectrum sensing of noise, Long-Term Evolution (LTE), New Radio (NR), and radar through deep learning. The following methods are used: a) different Cellular Neural Network (CNN) models such as resnet18, resnet50, mobilenetv2, etc., b) modifying segments of code, c) manipulation of the dataset. For the results, the modified normalized confusion matrix values significantly increased compared to the pre-modified version. For the spectrogram sensing, the received spectrogram, true signal labels, and estimated signals were dominated mainly by noise. Based on the graph, the frame mean Intersection over Union (IoU) has noticeably increased; however, based on data metrics, the mean IoU decreased. There is an increase in global accuracy, mean accuracy, Weighted IoU, and Mean BF Score. In conclusion, the objective of increasing the accuracy rate was met through various modifications, such as using different CNN architectures, parameter adjustments, and manipulating the given data set. However, the decrease in the mean IoU should also be considered, as it is also part of the accuracy measurement.