The problem of object detection is relevant for the automation of various processes. Currently, methods of detection and neural network architectures continue to develop. The classical methodology of using neural networks involves training models based on training data. If it is necessary to adjust the model, it is retrained using new training data or other training parameters. In the case of several trained models, the problem arises of choosing the final model that will be used for detection. This paper proposes an alternative approach, which consists in using an algorithm that allows to detect objects based on the analysis of detection results obtained using several models. The algorithm is synthesized and software implemented in Python. Computational experiments were performed for tomato detection using models trained on the basis of the YOLOv8 neural network.