A Scab Detection And Location Technique Based On Stereo Vision

Nowadays in China, nearly 80% of pesticides are still applied extensively into the farmland. No higher than 20% of these pesticides are used, and the remaining 80% are left in soil and water, which could cause a serious waste and lots of severe environmental problems. It also affects the crops we eat every day. As an emerging pesticide application strategy, variant spraying technique allows us to take different actions to different areas of the farmland, according to their regional differences. Through this approach, it could enhance the utilization rate of pesticide, as well as protect the environment.According to the related researches, a scab detection and location technique based on stereo vision was proposed in this paper, and a detection system based on this technique was built. The operation mode of the detection system was as follows:Firstly, the spatial coordinates of the scabbed plant are acquired by the stereo vision unit aboard a mobile platform; Based on the spatial coordinates, all joint variables of the robotic arm could be acquired through inverse kinematics and used to move the robotic arm to a certain position; then a clear image of the scabbed plant is acquired using a CCD camera carried by the robotic arm, by image processing techniques, we could eventually acquire the proportion of scabs; at last, a spraying prescription based on this proportion is sent to the variable spraying unit.The main research works and achievements are as follows:1) The technology roadmap of this scab detection and location technique was illustrated, as well as its operation mode. Also the key function units of the system built based on this technique were discussed.2) The theory and related algorithms of stereo vision were studied. Based on these studies, a simple stereo vision unit was constructed. Then the system parameters were calibrated using Zhang Zhengyou calibration method.3) Some works had been done learning different stereo rectification and stereo matching algorithms. After the distortion correction and stereo calibration, the stereo vision unit was used to realize the 3D reconstruction and spatial location of target plants, through feature extraction and image matching algorithms. Three different image matching algorithms provided by Open CV, including blocking matching(BM), semi-global block matching(SGBM) and variational matching(VAR), were studied and compared, results showed that block matching performed the best, with the best matching effect as well as the least processing time. Eventually, a experiment was carried out to verify the location accuracy of the system, which proved that the relative location errors could be under 1.65%.4) The theory and algorithms relating to the modeling and Kinematics of robotic arms were studied. Then the D-H model of a robotic arm with 4 degrees of freedom was established. Based on this model, the relationship between its joints and connecting rods could be analyzed easily. In the MATLAB software, the modeling and simulation of the robotic arm was realized, as well as its operating range.5) Some research has been done on the extraction of scabs from the plant. Some image processing algorithms including gray processing, image enhancement, threshold segmentation and morphological transformation were studied, and their effects were compared. Experiment showed that after these managements, an image of the scabs was acquired, with a relative good quality. Through this way, the proportion of scabs was calculated eventually.The system built in this paper could be applied not only to realize variant spraying, but also to acquire various kinds of field information integrated with other field information acquisition devices.