| Apple grading is one of the important parts of apple’s post-harvest processing and marketing.At present,apples need to be transported to the designated place for grading after harvesting,which increases the probability of damage,and the loss due to damage can be up to tens of billions of RMB.Therefore,it is of great significance to study the online grading system of apple harvesting robot,which can reduce the operation link,improve labor productivity and reduce economic losses.To study the apple online grading system,the idea of hierarchical design was proposed firstly.Secondly,the hardness and tribological properties of apple were studied.Thirdly,the process of hardware design was stated and communication system based on CAN bus was established.Fourthly,the indoor investigation of apple online grading system was done.The main research contents and conclusions include:(1)The idea of hierarchical design based on the analyses of the traditional classification system,the structural features and operating characteristics of harvesting robot was proposed.The apple beyond the standard size was removed by the pre-grading mechanism so as to reduce the unnecessary workload and improve the classification speed.Two feasible schemes and their structures and working principles were introduced,the final design was confirmed after careful comparisons.(2)The software and hardware designs of the system were completed based on the study of the hardness and tribological properties of apple.The results show that the apple hardness near stem and calyx were larger than that of the middle part,the apple hardness decreased with the increase of the diameter of apple.The sliding friction angle and rolling friction angle between the apple and the rubber plate are 30° and 17°.The critical damage velocity of the apple was 1.53 m/s and the best installation angle of the pre-grading mechanism was set at 23°.Besides,the 10 mm thickness silicone foam board was choosed as cushioning material to avoid apple damage.The vision system used single camera with rotating fruit to collect the surface images of apple and the black fabric was selected as the photographic background cloth.The S type weighing sensor was used to obtain the weight of apple,the grading actor mechanism was established by using the stepper motor to transport the apple while the non-toxic polyethylene material was used as pipeline material.A distributed control network based on CAN bus technology which used standard frame as the communication frame format was build.The thought of multi-thread technology and modular design of the program for the host computer and lower computer software based on Visual Studio 2008 by calling MATLAB calculating engine was put forward to realize image processing.(3)The apple classification algorithm based on machine vision technology was studied.The hardware structure of image processing system was introduced firstly.Later,a detailed description of image acquisition method via Flycapture2,MATLAB and Visual Studio was done.The image after gray processing was filtered by Gauss filter,and then the gamma transform,the gray image segmentation based on the Otsu method and the morphological processing were used to extract the feature region by turn.The minimum external rectangle method,the gray transform and the pixel statistical method were used to extract the apple size information,the apple defection information and the apple rotten area calculation,respectively.Besides,the regression analysis method was used to establish the relationships between the identification size and actual size,the identification area and actual area.(4)The experimental investigation of the apple classification system was conducted in the laboratory.The results showed that the pre-grading mechanism could remove the apple beyond the standard size with 90.67%accurancy,and 92.58%of 65 mm apple and 93.33%of 70 mm apples could cross the grading hole successfully.The maximum absolute error between the measured and tested weight values was 2.91 g,the average absolute error was 1.05 g,the maximum relative error is 0.77%and the average relative error is 0.38%.The results of apple size tests showed that the maximum absolute error between the measured and tested values was 2.68 mm,the average absolute error was 1.16 mm,the maximum relative error is 3.49%and the average relative error is 1.45%.The results of apple defect tests showed that the maximum absolute error between the measured and tested values was 0.38 cm,the average absolute error was 0.11 cm,the maximum relative error is 17.39%and the average relative error is 7.12%.The classification success rate of the overall grading system could reach 89.71%while the average continuous grading time was 2.89 s.The classification system is stable,easy to expand,and possess high classification efficiency and accuracy,which could meet the real-time classification needs of apple harvesting robot. |
PDF Full Text Request