Research On Detection Method Of Citrus Element Deficiency Leaf Phenotype Based On Machine Vision

Deficiency disease is a phenomenon of abnormal growth caused by insufficient nutrients during plant growth.Currently,due to the lack of scientific fertilization management.The phenomenon occurs commonly in citrus orchards in China.This seriously affects various life activities of citrus plants such as growth,development and fruiting.Therefore,it is important to establish accurate and convenient nutrient element diagnostic methods for crop growth management.In this paper,we propose a machine vision-based method for citrus deficiency detection based on the characteristics of the leaves of navel orange in southern Jiangxi and tangerine in Taizhou as the research objects.The main research contents and experimental conclusions of this paper are as follows:(1)Design and construction of a specific image acquisition device.In this paper,we investigate the differences in shape,color and texture of citrus leaves that exist due to different nutrient deficiencies.Due to the differences in their light absorbance in different spectral bands,a comprehensive spectral image acquisition scheme for citrus leaf reflection and transmission was designed.Meanwhile,this paper determines the selection of each component in the device through theoretical calculation and designs the installation scheme of the hardware structure.The device components specifically include camera,lens and light source.Through preliminary testing,the illumination scheme of the device is determined,and the full-band strip light source pair is used for low-angle front light illumination,and the 660 nm wavelength light source is used for backlight illumination.And the optimal light intensity under the reflected light and transmitted light conditions was explored.The results show that when the intensity of reflected and transmitted light is 92 and 70,respectively,the acquisition is uniformly illuminated without overexposure.Under these light conditions,the leaves were placed in the collection area and the leaves were laid flat with low-reflective glass.Reflected light spots in the leaf area were eliminated by using a polarizer.(2)Design image pre-processing and feature extraction algorithms.The spectral information from reflection and transmission is analyzed for their respective feature performance capabilities.First,since the color temperature of the light source causes bias in the camera’s acquisition of the actual color,this paper utilizes polynomial regression for correction.Second,under the B component of the image,the image segmentation of the reflection and transmission image pairs is performed using the OTSU thresholding segmentation method to obtain the target image containing the leaves.Under the H component of the image,the feature regions are separated using the threshold segmentation method.The preprocessing results in three types of target images with uniform size of 1024 ×1024,including the reflected leaf image,the transmitted leaf image and the feature leaf image.Finally,the target images are feature extracted in terms of shape,color and texture,and a feature dataset of citrus deficient leaves is formed after adding type labels,which vectorizes the image representation.(3)Build citrus leaf nutrition deficiency diagnosis model.In this paper,SVM classification training is performed for each of the three target images,and the accuracy of the model after different feature vector combinations is explored using the grid search method and the hierarchical cross-validation method,and the classification effects of the feature vector subsets after feature selection(RFECV)are compared.The results show that the model after performing feature selection has a good recognition effect,the test set maintains a high accuracy in the model,and the dimensionality of the feature vectors is reduced.Among them,the recognition accuracy of the model built by combining the feature leaf images with reflection and transmission information reached 95.20%.(4)Develop citrus leaf nutrition deficiency diagnosis software.This paper designs and develops the application software for citrus deficiency detection system based on Py Qt5 and Qt Designer interface design tools.The human-computer interaction interface with simple interface and easy operation is designed in combination with the image acquisition device.It can realize the acquisition of the image information of the leaf to be tested and the identification of the deficiency type.It supports the expansion study of custom sample types and quantities.