Applications Of Anomaly Detection In Detecting Pesticide Points In Vegetables By Using Hyperspectral Imagery

At present,the possibility of vegetables containing pesticide residues is increasing,and the relative mature pesticide residue detection technology such as gas chromatography has obvious shortcomings of timing-consuming,strong destruction.Therefore,there is an urgent need for a fast,nondestructive technology of detecting pesticide residues in vegetables.Hyperspectral imaging technology can distinguish the slight difference of spectral spectrum of ground objects,which has been applied to the field of food safety monitoring.Up to now,some achievements have been made by using hyperspectral imaging technology,and it has a big potential in the field of detecting pesticide residues in vegetables.The anomaly detection is the most common used target detection method in the field of hyperspectral remote sensing,especially in military target recognition,camouflage target differentiation and so on.In view of the effectiveness of anomaly detection in target recognition in hyperspectral remote sensing,this paper applies it to near hyperspectral images and tries to explore the feasibility of detecting pesticide points on vegetable leaves.Firstly,this paper set up a near-sight hyperspectral imaging platform to obtain the spinach hyperspectral image data containing pesticide points.And spectral characteristic of spinach leaves and pesticide points of different morphologies were then analyzed in order to determine the characteristic band length of pesticide spectral.Secondly,Singular Spectrum Analysis(SSA)in the field of signal processing was introduced to reduce the noise of hyperspectral image data obtained by experiments,which were affected by the photoelectric effect of the hyeprspectral imaing system.In order to evaluate the denoising effect,local image variance and four spectral similarity measurements were employed.The experiment results showed that compared with the original data,the local image variance of the image data after SSA treatment was reduced by more than 70%,and the four spectral similarity indexes had also decreased in different degrees,which effectively reduced the system noise and improve the data stability.Thirdly,for the problem of pesticide point detection,based on the basic principle of RX algorithm,the outer-window of double-window local RX algorithm was improved to be variable,and thus combining with Principal Component Analysis(PCA),a method of data dimensionality reduction,an algorithm named variable outer window local RX algorithm based on principal component analysis(variable outer-window PCA-LRX)was proposed.Experiment results showed that compared with other four RX series algorithms,the variable outer-window PCA-LRX algorithm achieved the best detecting effect,which could detecting all the pesticide points with very low false alarm rate.The Receiver Operating Characteristic curve(ROC)indicated that the variable outer-window PCA-LRX algorithm achieved the highest detection rate under the same false alarm rate.Finally,to solve the long running time problem of the variable outer-window PCA-LRX algorithm,based on the basic principle of Hausdorff distance,combined with PCA,an improved Hausdorff distance anomaly detection algorithm,namely PCA-MHD algorithm was proposed.Experiment results showed that the PCA-MHD algorithm could also detect all the pesticide points,and compared with the variable outer-window PCA-LRX algorithm,the running time was greatly shortened with high detection efficiency,which provided a new idea for the detection of vegetable pesticide points.