Prospective of Camouflage Ground Objects to Counter UAVs

Unmanned aerial vehicles (UAVs) are becoming increasingly important for military operations, including reconnaissance, providing battlefield support, and performing a variety of missions. The possibility of attacking targets using a group of UAVs increases the relevance of protection measures, particularly camouflage. The literature review highlights the experience of using UAVs in military spheres of activity and their high efficiency in reconnaissance and support on the battlefield. Using high-resolution images, visual orientation on the terrain was considered a priority direction of UAV navigation in the operation of electronic warfare tools. Based on hybrid convolutional neural networks, hardware for working with the UAV platform is serially produced. Vines are suggested for masking stationary objects due to their rapid growth and the corresponding quality/cost ratio. The work aims to check the effectiveness of masking small objects using grassy and climbing (lianas) vegetation concerning the prospects of their identification by optical means. The research was conducted using our data obtained from UAVs in the visible range on experimental fields and the botanical hall of NUBiP of Ukraine and on pictures of buildings from the Internet. It has been experimentally established that vegetation's variety of shapes, sizes, and structures complicates the recognition process for convolutional neural networks. The appearance of plants can change significantly depending on the angle of photography, lighting, and the state of mineral nutrition. Organic camouflage in the form of grasses and vines can entirely or partially hide buildings and structures, changing their brightness and contour. This can obscure details and increase the risk of false object recognition. For example, facilities covered with ARISTOLOCHIA MANSHURIENSIS Kom vines were correctly recognized only 27% of the time using neural networks. Convolutional neural networks require a large amount of data to train if the goal is to achieve high accuracy. However, if data with vine-covered objects is limited or unavailable, the training process may need to be improved, affecting the network's ability to recognize such things accurately.