Research On Soybean Pod Morphology Classification Method Based On Machine Vision

Soybean is one of the main crops in China and even the world.It is also one of the important and strategic crops in China.Because it can be used for both grain,oil and feed,it has excellent application value.At present,soybeans grown and produced by our country are far from fully able to meet domestic needs.The large demand for soybeans has promoted the selection and breeding of excellent soybean varieties,thereby enhancing the overall quality of soybeans and providing assistance for increasing domestic soybean production.The morphology of soybean pods is one of the important reference indexes for breeding high-quality soybeans.Therefore,how to accurately classify the morphology of pods is very important.Traditional methods generally use manual methods for classification.This classification method depends on the individual's past classification experience and their own visual observation ability.The classification speed is very slow,which consumes people's energy seriously,and the classification process will be affected by subjective factors.It is easy to cause the c lassification criteria to fluctuate,making the classification results inaccurate.In order to overcome the shortcomings described above and make the classification work efficient,objective and accurate,this paper proposes an automated classification method based on machine vision.The experimental results show that the method proposed in this paper can be used to automatically classify the morphology of soybean pods,which has a good classification effect.This method also has good potential development in other classification tasks for the soybean field,such as soybean leaves,grains and diseases,etc.The specific work content includes:(1)This paper designs an image acquisition system.The image acquisition system does not use the industrial cameras commonly used in general methods,but uses high-performance digital cameras as the core,combined with flexible and adjustable LED light sources.The image acquisition system can obtain high-quality soybean pod images in batches.(2)This paper designs an image preprocessing method,which can extract multiple pods from each original soybean pod image data acquired by the image acquisition system and perform normalization operation.Using this method,all the original soybean pod image data collected in this paper is made into a data set with a uniform format and style,which is convenient for the subsequent classification work.(3)This paper designs a method for morphological classification of soybean pods based on traditional features.This method extracts a series of traditional features manually from the pod images after the preprocessing operation.These features are used to train the support vector machine classifier to classify soybean pod morphology.The classification accuracy of the automatic classification for straight,curved and bow soybean pods is 95.2%,96.6% and 93.3%,respectively,and the average classification accuracy can reach 95.2%.(4)This paper improves the classification method in(3),and designs a morphological classification method for soybean pods based on deep features.This method designs a convolutional neural network model,which is used to extract the deep features of the pod image.Deep features are used to train support vector machine classifiers to achieve more accurate classification of soybean pod shape.The classification accuracy of this method for straight,curved and bow soybean pods is 98.8%,99.2% and 98.1%,respectively,and the average classific ation accuracy can reach 98.7%.The experimental results show that the classification a ccuracy using SVM with deep feature training is increased by 3.5% compared with those with traditional feature training,which proves the advantage of using deep feature training.