The main goal of this research is to determine the level of surface contamination in a considerable area of land covered with water, such as the Colta Lagoon, through the implementation of recognised use drone cameras, by integrating machine learning techniques with computer vision, which can detect objects, track various materials, and classify them accurately in real-time or not, allowing the determination of the type of pollutants existing in these experimental spaces after analysis, with autonomous flight capabilities. The system comprises a convolutional neural network (CNN) and an AlexNet network that has already been trained. The objectives are to achieve near real-time performance and the highest accuracy possible by building a database of specific images with transfer learning terms. It is therefore determined that the accuracy of the system is extraordinary both under real conditions and when using a standard dataset. This high rate of precision in applying the proposed model to real data leads us to develop a method for use in video-based artificial vision applications.