| Studies on the key techniques of the fruit-picking robot are the necessary circular to large-scale facility agriculture and industrialized process. Studies are made for the operation information acquization of tomato-picking robot, and it is of great theoretical significance and practical values to improve the research and development level of intelligentized facility agriculture in China.As for guiding the tomato harvest robots harvesting selectively, issues that acquire information quickly and accurately have been studied. In order to obtain the information of ripen fruits grades for picking and storing tomatoes can be classified with real-time, thus the detection has been conducted with multi-sensor infusion of computer vision images and reflected near-infrared spectra. And the path for manipulator picking multi-objects has been planned. In this paper major research works are as follows:1. A method of test information collection has been studied for image, spectroscopy of tomato and normal detection of tomato internal composition. Based on the filtered in the maturity of the tomato testing samples, the information acquisitions are been discussed for visible image, multispectral images of tomato and road image in greenhouse. The testing scheme about acquisition of tomato near-infrared spectroscopy was introduced with the spectrometer. The generalized detection indexes of tomato internal physical and chemical composition are been conducted by normal detections and process.2. By using machine vision technology, the tomatoes’ maturity has been identified under natural growth conditions of the field. The tomato samples chosen as research objects are divided into five different stages, such as breakers, turning, pink, light-red, and red stages. The visible light and near-infrared images of tomatoes have been acquired. And12color eigenvalues of images were extracted and analyzed. Tomatoes are classified with using pattern recognition methods, such as neural networks and decision analysis. The results indicate that:(1) with the changes of maturity, the Hue-means and Green-mean of images decreased gradually, and the standard deviations of the images’Hue-mean and Green-component mean are the largest values for tomatoes in the pink stage.(2) The intensity mean of tomato samples’near-infrared images in the pink stage is the lowest value.(3) Hue-mean can be used as a criterion for judging tomatoes’maturity. When Hue-mean taking43, tomatoes can be divided into maturity of breakers and turning above (including turning).(4) Based on the statistical model of Hue component, the accuracy of determination for tomatoes’maturity model is93%, and the judgment errors are caused by identifying tomatoes at the pink stage.3. Rapid detection has been studied for the internal quality of tomatoes based the techniques of visible-near-infrared spectroscopy. In order to detect the internal quality grades of tomatoes during harvesting selectively, the non-destructive rapid detections are taken by using the techniques of visible-near-infrared spectroscopy against carotenoids, solids, total sugar and total acid content of tomatoes.(1) A total of70tomato samples produced in Zhenjiang are chosen as the research objects. Three different spectral pretreatment methods are compared on the accuracies of the regression mode. Four physical and chemical compositions of support vector machine regress models have been established in the range of full wavelength with normalization of sampling data set. The cross validation correlation coefficient of the model is over0.94.(2) Sugar is selected to evaluate the intrinsic quality of tomatoes. By using iPLS method,216wavelength points are extracted from the full-waveband range, and further6characteristic wavelength extracted by using genetic algorithm in combination with partial least squares (GA-PLS) are522nm,557nm,1215nm,1251nm,1279nm, and1284nm, respectively. Based on the extracted6characteristic wavelengths, support vector machine classification models have been established for judging tomatoes’sugar grades. The accuracy of classification is82.35percent. And sugar can be used as an assessment indicator for rapid detecting tomatoes’quality.4. Studies are made for rapidly identifying and extracting the tomatoes sheltered under the condition of a greenhouse. The Hue component of the color characteristics is proposed to segment the sensitive area of the tomato. For recognizing the area of tomatoes, the background noises of the leaf, stem and greenhouses’appendages are eliminated by the morphological opening operation method, and bright spots caused by direct sunlight are eliminated by the method of filling regional hole, and the marked regions are dealt with the method of sequence for tracking the boundary of multi-tomatoes areas. The shape characteristics of multi-tomatoes seriously overlapped are extracted in a variety of natural growth conditions. Characteristics parameters of shape in each tomato, such as the center point and circumradius, are determined based on the cluster algorithm of the boundary string bisector. The results show that the processing time of the image with640X480is0.45second, and correct recognition rate for200images of tomatoes under the natural state is95.5%.5. In the period of harvesting, the identification of the ropes fixing the tomatoes’ stems are studied. At first, the Otsu method is used to determine the threshold segmentation of the rope images, and a modified amount of threshold suitable for variety weather is determined by experimental analysis. Considering a variety of complex noise interference of background conditions in a greenhouse, a threshold-based method is used to remove the noise region of the small area. And aspect ratio threshold method of external rectangular for the region is used to remove the area with large background noise for recognizing the rope area. Finally, the location of the ropes is determined using the method of least squares. Aspect ratio’s threshold value of circumscribed rectangular is6.0952determined by the experiments. The identification method has adaptability to interfere factors, such as the uneven light and complex background noise in the greenhouse. The correct identification rate of100sample images with obstacles is93%, and the average elapsed time is0.8second.6. System integration of multi-information acquisition has been studied for tomato harvesting robots. In the process of real-time classification harvesting, the tomato picking robot is explored to treat the selection of fusion hierarchy and information fusion composite structures with multi-sensors, thus making the achievement of selective harvesting decisions and grade determination of tomatoes based on the multi-sensors information. A system of multi-information acquisition has been studied for guiding tomato harvesting robots in greenhouses. The binocular visual recognition system has been established. The information of fruits’sugar grades is acquired by the FieldSpec Pro spectrometer. The tomato picking manipulator is composed of a MOTOMAN manipulator accompanied by a self-developed end-effector. And the tomato picking manipulator and a movable platform constitutes an8-DOF tomato harvesting robot. It has been known through calibration experiments that the reasonable mounting pitch of the binocular cameras is analyzed to determine it as250mm, and positioning tolerance can be controlled within5mm. Based on the mixed programming of Visual C++and MATLAB, modular design has been used to develop the software of the harvesting decision-making and target information detector, and then the testing detection is made for the selected samples.7. In the process of real-time classification harvesting, the complex path planning has been studied for the manipulator going from muti-fruit location points to multiple boxes of fruits. A method that builds a manipulator global optimal path planning decision tree is proposed, thus searching the decision tree comprehensively according to the optimal principle of the global shortest distance so as to establish path optimal model and to solve the global optimal path. Combining the spatial location of tomato fruits with the location information of hanging vine rope obstacles, the kinematics model has been constructed for MOTOMAN SV3X manipulator, and studies are made for local obstacle-avoiding movement trajectory planning.8. A method has been studied that identifies the central path of greenhouse operating environment with the non-structural and complex background. A method has been proposed that segments the image adaptive threshold with maximum interclass variance according to the I component histogram, and obtains heat pipe edge discrete points cluster by using the edge extraction algorithm of the target area after the segmentation of binary image, and then obtains the two heating tube edge line after its bridging the method of least squares. Based on this, a baseline extraction algorithm has been derived for the center of the road, and this road detection algorithm is verified for uneven illumination and crop occlusion factors. The results show that the average deviation between baseline of the center of the road used with this method and the manual fitting method is0.77%, when the heating pipe shadow rate is between10%and90%, with91.4%of the accuracy of road baseline extraction algorithm, and with0.72%of the average relative deviation. Especially, this average time is76.4%lower than the Hough transform for line detection algorithm, which shows that this algorithm is simple and quick with satisfactory robustness. |
PDF Full Text Request