| The seasonal strength of apple harvest results in a tight labor supply and the cost of artificial harvest is increasing year by year.Scientists attach great importance to the research of mechanized harvest of apple,and the pre-grading classification treatment after apple harvest can greatly reduce costs and increase efficiency.In this study,the Fuji Apple harvested in the field is processed in real-time classification,and its appearance quality such as size,color,shape and surface defects are comprehensively detected.The key technical problems such as the development of small equipment applied in the field,the real-time detection of various fruit surface defects and the restoration of apple dynamic image are solved.The main contents and conclusions are as follows:(1)This paper analyzes the development status of apple grading equipment at home and abroad,and puts forward the existing problems in the research and development of apple grading equipment by comparing the differences between field grading and industrial grading.According to the planting mode of wide row and short anvil in moder n orchard,the overall scheme of Apple field grading equipment was determined.The prototype has a total length of 1500 mm,a total width of 400 mm and a total height of 1000 mm.The grading system is driven by a speed-regulating motor to carry out the automatic grading of apples harvested in real time in the field.It is small and suitable for field transfer and turning.Jeston TX2 artificial intelligence computing platform is selected as the processor to meet the requirements of computing power.(2)The key parts were designed and the prototype assembly was completed.According to the measured apple appearance characteristic parameters and relevant theoretical knowledge,combined with CATIA three-dimensional design software to guide the mechanism,transmission device,actuator and other key parts of the design.Among them,the actuator is composed of three parts: signal detection device,driving device and terminal actuator.The parameters of the parts and components that meet the requirements were calculated and selected.Finally,the prototype was assembled to verify the rationality of the equipment.The results showed that the devices could cooperate with each other to complete the apple grading work in the field automatically according to the process design.(3)Based on machine vision and deep learning,the algorithm of Apple appearance feature detection is studied.The detection device is built to extract the apple image,and the deblurgan method is used to recover the dynamic blurred apple image in order to increase the stability of parameter extraction.According to the national standard of apple grading,the four appearance features of apple to be tested are defined,namely,size feature,color feature,shape feature and surface defect.According to the different features,the corresponding detection algorithm is designed separately,and the information fusion is carried out based on the specific data of these four features.Using the support vector machine method to identify,the apple is divided into three grades: first-class fruit,second-class fruit and outer fruit.Develop the apple detection interface synchronously,monitor the grading process in real time,and make the detection result more intuitive.(4)The accuracy of apple grading algorithm was tested and analyzed,and the apple grading equipment was verified in the field.The results showed that the accuracy rate of apple size detection was 99.04%,shape detection was 97.71%,color detection was 98.00%,and surface defect detection was 95.85%.The accuracy rate of first grade fruit,second grade fruit and other fruits was 93.24%,94.25% and 100%,respectively,with an average accuracy rate of 94.12%.Under the condition of the conveyor belt speed of 0.4m/s and the walking speed of grading equipment of 0.5m/s,the grading efficiency of apples is 40 /min,which can meet the requirements of real-time grading of apples in the field. |
PDF Full Text Request