Study Of Determination Of Activity For Antioxidant Enzyme System Of Tomato Leaf And Plant Based On Hyperspectral Imaging Technology

Agriculture is the foundation of national economy, and its development is closely related to the people's livelihood. With advance in technology, precision agriculture and internet of things (IoT) in agriculture have been continuously developed, which provide technical support for sustainable development of agriculture in China. The traditional means of chemical analysis and information retrieval for farming are unable to meet the need of development of modern agriculture and application of IoT in agriculture. Therefore, Rapid non-destructive testing technology is an area urgently needed developing. Because of traditional agricultural farming and increasingly serious environmental pollution, Heavy metal pollution has caused more serious effects on agriculture. In this study, tomato plants were as the research object and Cu2+ ion stress was as the environmental control condition. Antioxidant enzyme activities in tomato leaves were measured rapidly and non-destructively using hyperspectral imaging technology. Variation for antioxidant enzyme system activity of tomato leaves under Cu2+ ion stress were obtained. Rapid determination for activities of CAT, POD and SOD were achieved. Possibility for modeling and forecasting activities of CAT, POD and SOD based on texture parameters of hyperspectral images was discussed. Prediction accuracy of enzymes for whole plant scanning with application of prediction model based on standard scanning was evaluated. There were significance for rapid determination of tomato plant growth condition, rapid determination of enzyme activity for agricultural workers and sophisticated management of tomato cultivation. The main creative results were achieved as follows:(1) Under Cu2+ ion stress with concentration of 0 mg/L,25 mg/L and 50 mg/L, enzyme activities of CAT, POD and SOD in tomato leaves increased with the increase of stress concentration, and maximum values were 280U/g,566.38U/g and 290.91U/g, respectively. Differences in enzyme activities for CAT, POD and SOD under different Cu2+ ion stress concentration were chemical foundation for modeling prediction.(2) Activities of CAT, POD and SOD in tomato leaves ware measured form spectrum dimensional analysis of hyperspectral images. Operation process was:Extraction of spectraâ†'Pretreatmentâ†'Extraction of characteristic wavelengths using SPAâ†'establishment of prediction model. The optimal pretreatment was DOSC among the six methods such as SG, SNV, MSC, 1-Der,2-Der and DOSC. CAT activity prediction models shown that SPA-PLS model in all such as SPA-MLR, SPA-PLS, SPA-BPNN and SPA-LS-SVM based on reflectance information of effective wavelengths (958nm and 419nm) extracted by SPA was the best effective model for forecasting CAT activity with Rp and RMSEP being 0.9800 and 12.12U/g, respectively. POD activity prediction models illustrated that SPA-PLS model in all such as SPA-MLR, SPA-PLS, SPA-BPNN and SPA-LS-SVM based on reflectance information of eight effective wavelengths (443nm,464nm,413nm,410nm,401nm,402nm,426nm and 926nm) extracted by SPA was the optimal effective model for forecasting POD activity with Rp and RMSEP being 0.9353 and 37.80U/g, respectively. SOD activity prediction models demonstrated that SPA-PLS model in all such as SPA-MLR. SPA-PLS, SPA-BPNN and SPA-LS-SVM based on reflectance information of three effective wavelengths (978nm,401nm and 418nm) extracted by SPA was the optimal effctive model for forecasting SOD activity with Rp and RMSEP being 0.9476 and 19.28U/g, (?)tively. The results indicated that prediction accuracy of models were satisfactory for CAT, (?) SOD activities using spectral-dimensional models combined with chemometric methods.(3) In image-dimensional analysis, using texture parameters of hyperspectral images to pied(?)t activity of antioxidant enzymes in tomato leaves was studied exploratory. MNF was used to process the original hyperspectral images. PCA was applied to extract PCs images. Seven effective wavelengths were extracted by loading factors of PCs images and eight texture parameters were obtained from each PCs images and gray of effective band. Four prediction models including PLS, DOSC-PLS, PCA-BPNN and PCA-LS-SVM were established using texture parameters. The best effective model for prediction of CAT activity was PCA-BPNN with Rp and RMSEP being 0.6830 and 35.85U/g for prediction set, respectively. The optimal model for pr(?)on of POD activity was PCA-LS-SVM with Rp and RMSEP being 0.6894 and 43.08U/g for prediction set, respectively, Rp of prediction set for the rest of prediction models was less than (?). The optimal model for prediction of SOD activity was PCA-BPNN with Rp and RMSEP being (?) and 45.84U/g for prediction set, respectively. Rp of prediction set for PLS and DOSC-PLS (?) than 0.5. Non-linear model PCA-BPNN was significantly better than linear model. Conclusion mentioned above represent that prediction accuracy using texture parameters to establish model for predicting the activity of CAT, POD and SOD in tomato leaves was very poor. (?)vas (?) work for model improvements to increase difference among enzyme activities, (?)important texture parameters associated with enzyme activity.(4) Distrabution of POD activity in leaves was drawn. DOSC-SPA-PLS for CAT prediction model in standard scanning was employed to predict CAT activity of tomato leaves under whole plant scanning. The result showed that, although prediction accuracy of DOSC-SPA-PLS for stan(?)ng was excellent, for whole plant scanning was unacceptable. The difference (?)values and reference values was apparent. This proves that DOSC-SPA-PLS (?) standard scanning was not applicable to whole plant scanning for CAT activity (?) (?)lusion was drawn that establishment of CAT enzyme and other antioxidant (?)nzyme activity prediction models under whole plant scanning was based on re-establishment in (?) plan(?)ing or calibration of model obtained in standard scanning. This is difficult task.