Research On Fresh Peach Bruise Detection Based On Hyperspectral Image

Fresh peach is a seasonal fruit,which is favored by customers because of its thin,juicy pulp and sweet taste.However,fresh peaches are easily bruised during transportation which will affect their quality.Therefore,fast and efficient detection of fruit bruises is of great significance to the storage,transportation and process of fruits.Traditional methods of detecting fruit bruise are incredibly destructive and laborious.To solve the problems of low efficiency and high cost of traditional detection methods,the hyperspectral "map integration" technique was used to detect bruises on fresh peaches in this study.The research objects included white peach,honey peach and flat peach.First,the hyperspectral feature extraction algorithm was used to select the feature bands.Then,the corresponding characteristic hyperspectral images were extracted according to the characteristic bands.Furthermore,the image fusion algorithm was used to fuse multiple characteristic hyperspectral images.Finally,it is feasible to detect fresh peach bruises based on hyperspectral.The research results are as follows:(1)Research on the difference of spectral reflectance of fresh peach samples with different bruise degree.First of all,hyperspectral images of white peaches,juicy peaches and flat peaches with different degrees of damage were obtained by using a hyperspectral instrument.Secondly,the hyperspectral reflectance of three fresh peaches based on the region of interest was extracted and the corresponding spectral curve was constructed.Finally,it was found that the spectral reflectance of non-bruised part was higher than that of slightly bruised parts,and that of slightly bruised part was higher than that of severely bruised part by analyzing the spectral reflectance of fresh peach of the same variety.In addition,the spectral characteristics of white peach,honey peach and flat peach were similar.(2)Research on the fusion of characteristic hyperspectral images.First,the hyperspectral analysis was performed on fresh peach samples with different bruises.Secondly,two characteristic bands(3 channels and 10 channels)were selected based on the successful projection algorithm(SPA).Finally,two methods including the pixel weighting method and wavelet transform method were used to fuse the 3-channel and 10-channel characteristic hyperspectral images,respectively.The results of the experiment showed that the fusion effect of 10-channel feature hyperspectral images based on wavelet transform method was better.(3)Research on fresh peach bruise detection with fusion feature hyperspectral images.10-channel images were fused based on wavelet transform as experimental data,the OTSU,K-means,segmentation algorithm based on Niblack local threshold and Pyramid Scene Parsing Network(PSPNet)were used to detect bruises of different fresh peaches.The results of the experiment showed that different methods have different detection accuracy.Among them,the detection effect of fresh peach bumps based on the PSPNet model was the best.For white peach,honey peach and flat peach,the detection accuracy were 0.953,0.959 and0.948,respectively.