Farmland roads and ditches are the majority of land consolidation project and construction of a high standard of basic farmland, which can be monitored by the high spatial resolution remote sensing. An approach of automatic linear project feature extraction with WorldView—2 high-resolution image based on Canny edge detection algorithm was presented, which was designed especially for field roads and ditches detection in the farmland. Firstly, based on the Canny edge binary images, single-pixel gap was connected by matching one of eight masks summarized. Secondly, the suspect linear feature was identified by Euclidean distance judgment. Each pixel of enhanced images was scanned with the seed growth method. Thirdly, each line was joined into the intact linear project feature with a lower length threshold. Fourthly, redundant features were deleted with a higher length threshold. The steps listed above were divided into two directions including vertical direction and horizontal direction and the final results were formed by combining results from the two directions. Experiments were designed in the land consolidation area of Dalad Banner, Inner Mongolia. The accuracy assessment was presented, the precision was above 95% and the correctness in the simple conditions was above 95%, which also met the basic needs of land consolidation project monitoring. Based on Canny edge detection, the algorithm of this paper can be realized in a fast and efficient way. All the procedures of this algorithm were realized automatically by adjusting operators without requiring artificial interpretation.