Content area
The aim of this study is to develop a convolutional neural network architecture designed for apple recognition in images. The relevance of this task is tied to the need for fruit recognition to automate the process of apple crop harvesting. To reduce computations, it is proposed to convert the image captured by the camera from RGB format to HSV format. Using the example of a red apple, the creation of a bitmask is demonstrated, which allows for the identification of regions of the desired color within the image. A structure and parameters of the convolutional neural network were proposed, along with a method for computing the distance between the detected object and the camera based on the pre-calculation of the focal length. To analyze the results of the neural network under consideration, software was developed in Python using the TensorFlow and Keras libraries. The training and testing of the neural network were conducted on a PC Aspire A315-23 with an AMD Athlon Silver 3050U 1.2 GHz processor, 4 GB DDR4 RAM, and an AMD Radeon Graphics 2.30 GHz graphics card, running Windows 11 Pro operating system. The neural network was trained for 15 epochs, taking 217 seconds in total. Object recognition by the trained neural network took around 1 second. The proposed convolutional neural network model demonstrated a recognition accuracy of 86% on the test image set.