Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Abstract. Autonomous traveling which is one of the essential components for agricultural robot is expected to be efficient for farming automation in that the agricultural machinery travel to work along the crop row repeatedly. In this paper, the practical and efficient framework for autonomous path detection in semi-structured environment, orchard was proposed. The framework can segment area in big size images of frontal scene, and it is expected to apply various kind of area detection tasks in agriculture fields. The main point to this process is deep-learning-based path area localization, and it consist of cropping the image patches from input image using sliding window, generating path score map using CNNs-based classification between tree and path, path area estimation, and target path detection using boundary lines determination. To implement and evaluate the framework, path detection for autonomous traveling in citrus orchard was conducted using the developed technique. Real-time images were taken using RGB camera attached on the speed sprayer with remote control manually and captured images could be classified 4 cases. The results showed that the averaged errors were observed 0.056 and 8.1º for lateral and angular difference, respectively. Based on the results, the proposed framework can perform as well as other approaches including laser scanner, GPS, and vision sensor although it needs low cost memory to execute the framework compare to other deep learning-based frameworks like object detection and segmentation. It is possible that the autonomous farm robot is developed using this proposed technique easily with simple hardware configuration.