Research And Implementation On Automatic Method Of Apple Detection And Phenological Period Classification

The process of China’s agricultural modernization continues to advance.Apples are the main cash crop in many places.Increasing the classification accuracy of the phenological period of apple trees and the detection accuracy of fruit trees are conducive to improving the scientific and automatic level of apple tree planting and the economic benefits of the apple planting industry.In actual production,the classification of apple phenology mainly depends on labor,while in scientific research,there is less research in this area.Therefore,this paper proposes an automatic classification method for apple phenology based on convolutional neural network.In addition,due to the complex background and unstable environment factors of apple tree images collected in orchard,there are a large number of overlapping and incomplete apple targets.Combining deep learning with traditional methods,this paper proposes a method of fruit detection and grading by color and apple diameter.Based on the above model,apple phenology classification and fruit detection system are designed and implemented,which has practical significance and application value for improving the automation level of apple planting.The specific research contents of this article are as follows:(1)In view of the shortages of apple tree data sets in orchards,this paper collects and processes apple tree images of different phenological periods and establishes apple tree image data sets.First,the contrast enhancement algorithm is used to preprocess the image,and then the random cropping method and the information deletion method are used to increase the amount of data.Finally,the author marks apple’s location in the image and creates a VOC format annotation file.(2)This paper proposes an automatic classification method for the phenology of apple trees.First,a convolutional neural network is established to extract image features,and then a fully connected network is used to extract one-dimensional time feature vectors.The extracted phenology images are merged with temporal features,and then input into a fully connected network to classify the phenology periods.(3)This paper proposes an apple detection and classification method.First,the k-means method is used to cluster and analyze the apple to obtain the detection box anchor box in the data set.Second,the convolution network is established to detect the position of the apple based on the multi-scale feature map for the first time.Third,the image is converted in the laboratory detection box color space,then analyze the main component of apple color and extract the b* component and 1.8b*-l* component respectively.After that,the OTSU method is used to analyze its gray value,locate the optimal segmentation threshold for apple and background and binarize the image accordingly.The detection box is corrected twice according to the minimum circumscribed rectangle,which improves the accuracy of the detection box and the detection rate of small targets.Calculate the apple diameter according to the Image Size Conversion Model,and complete the apple grading according to the Apple Grading Standard.(4)This paper designs and implements an apple phenology classification and fruit detection system.The system includes user settings,registration login module,image import module,image normalization module,image phenology classification and fruit detection module,result query and analysis module.The system stores the raw data and detection results of apple tree images.The system is applied to the Apple Demonstration Park,which improves the automation level of apple cultivation.