The Defects Detection Based On The Apple’s Infrared And Visible Images’Fusion

As one of the biggest apple product countries, it is necessary for China to control theapple defects detection strictly to improve the quality of Chinese apples or apple products,increase the apple exports and strengthen the market competitiveness. At present, the defectdetection of fruits in our country has been predominantly performed by manual assessment,slice sampling, or visible spectrum imaging systems, that always have a much highmisjudgment rate and do not meet the requirements of fruits online detecting, which should benondestructive, rapid, and intuitive enough.Aiming at the shortcomings above-mentioned of the traditional defecting methods, theimage fusion technology of infrared and visible images was introduced in for the fruit defectsdetection. According to the specific situation of China, Red-Fuji apple was chosen as theobject, and the key technologies of apple defects detection were researched in detail in thispaper, mainly from the image collection and integration. Relative merits and demerits ofseveral common fruit defects nondestructive detecting technologies were firstly analyzed, andthen on the basis of this, the apple defects detecting method, which was based on infrared andvisible images’fusion, was finally determined. Besides, the main research contents were listedin the paper then.In the fore part of the paper, the theoretical basis of infrared and visible images’ fusionwas researched. First of all, the basic theory of infrared and visible images’ fusion, which wasmainly including the basic mechanism and characteristics of infrared imaging technology, thethree levels of image fusion, and the basic fusion methods. Then, the overall scheme of theapple defects detecting system was designed, as well as the design of the infrared imagecollection, including the choosing of incentive pattern, the placing of incentive tools, and thebuilding of surface thermal radiation model of the apple. After of that, several typical infraredimages of the apple was provided, as well as the corresponding characteristics’ brief analysisof them. In addition, lines and points of characteristic were selected in the thermal images, sothat the temperature and coordinates of the selected points were obtained, as well as thecurves of temperature changing along the selected lines.In the second half of the paper, the focus and key technologies of the detection systemwere researched in detail, which were mainly about the fusion of the apple images, includingimage preprocessing, image registration, and image fusion. The main contents of imagepreprocessing were about image denoising and background division. In this part, the methodsof median filter and wavelet denoising were introduced respectively for grain noise and whitenoise. And then, a method for background image segmentation was proposed in the paper, which was based on the threshold of the apple image gray levels. Characteristic points werepicked up based on the Harris operator, and the image registration experiment was carried outunder four different matching transformation models, by which the suitable transformationmodels for this paper was chosen then. The image fusion was achieved based on the modifiedwavelet transform:○1the discrete wavelet decomposition of the apple infrared and visibleimages;○2the adaptive weighted fusion of the low frequency wavelet coefficients based onimage regional energy;○3the adaptive weighted fusion of the high frequency waveletcoefficients based on the image regional strength.○4the reconstruction of the wavelet forgaining the fused apple images. In the last of the paper, the fused images under the abovealgorithms were provided, as well as the associated objective evaluation and experimentalverification. And finally, several involved research conclusions were summarized, proving thefeasibility and advantage effectively.