Research on Image-Guided Positioning Technology for Automated Agricultural Machinery Based on Convolutional Neural Networks (CNN) and Visual Servo Control

Automated agricultural machinery often suffers from low positioning accuracy and unstable target recognition in complex farmland environments. These problems affect path planning, operation precision, and the reliability of autonomous control, especially when visual perception must operate alongside wireless communication and electromagnetic sensing modules in harsh outdoor conditions. This paper addresses these issues by constructing an image-guided positioning model based on convolutional neural networks (CNNs) and visual servo control. The model acquires crop and environmental images through a multi-source image acquisition module, extracts features and recognizes targets by an improved lightweight CNN with an attention mechanism, and inputs the recognition results into a visual servo control system. Real-time position correction is performed together with depth-sensor feedback to achieve accurate positioning and path tracking. The method includes establishing a multi-scale image sample set, designing an attention-enhanced CNN, embedding proportional-integral-derivative adjustment into the visual servo loop, and correcting system deviations through an error-compensation model. Experimental results show that the positioning error remains between 2.4 cm and 2.8 cm in complex field environments, with an average response speed of 83.8 ms. The method therefore improves the automation, stability, and engineering applicability of image-guided agricultural machinery.