Freshness Detection Of Spinach Based On Hyperspectral And Imaging Procsess Technology

Spinaches which are rich in water,chlorophyll,mineral salt and so on are obbligato parts of the resident’s food.After spinaches were picked,a series of physiological changes will occur in the plants.The property of vegetables will descend.In order to achieve the most benefits,several illegal sellers regard the worse as the better.The behavior damagesbuyers’ life,health and profit.It goes against the benign market competition.Hence,it is essential to detect the freshness of spinaches.Currently,spinaches are discriminated mainly by manual work and traditional machine vision indeed.The former needs experienced experts full of subjective ideasand sometimes samples may be destroyed.The latter just pays attention to visible light images and will lose lots of crucial features,leading to worse precision.Attempt to solve above problems,the paper takes spinaches stored under room temperature 10℃ as objects of study.Based on hyperspectral and image processing technology,a number of methods will be taken to select waves.The research found that LBP feature of the 930.69nm image(f eaturebest)had the best recognition result and accuracy was up to 91.5%.The above study has realized rapid and undamaged detection of the spinaches’ freshness,providing quality discrimination guarantee for their process and market selling,and is meaningful practically and theoretically.The major content and innovation are acquired as follows:(1)Dividingthe spinach’s freshness level is a difficult and significant problem.Now,there is not the common industrial standard internationally.After normalizing the stored days,appearance,the content of water,chlorophyll a,b,carotene of every leaf,each index will get a score and standard deviation.On the basis of the standard deviation,indexes will be given weight.At last,samples will obtain total scores.Two break points(0.5208 and 0.3593)will be set as critical values.Sections of freshness,secondary freshness and corruption are respectively[0.5209,0.6676],[0.3594,0.5208],[0.031,0.3593].(2)Grouping Genetic Algorithm based on Elite(GGABE)is proposed to select hyperspectral waves.Two grouping methods are adopted:artificial grouping and adaptive grouping.Artificial grouping separated the whole population into 2-32 teams averagely;Adaptive grouping was determined by characteristics of the reflection.Waves which have the compact reflection and lower class-distance will be set as critical boundaries.After grouping,genetic operation will be conducted at every team with elite strategy independently.(3)K-Means Cluster Algorithm based on Adaptive Fish Swarm(KMCAAFS)is proposed to select best waves in the paper.Centers are generated randomly in the traditional k-means cluster algorithm and the contingency is much larger.In improved algorithms,two strategies are used:artificial fish swarm and feature-class correlation.Initial centers are determined by artificial fish swarm whose visions and steps are changed adaptively.Several individuals which have stronger classification capacity are selected as initial centers in feature-class correlation strategy.After determining initial centers,every individual will be allocated to the closet cluster.(4)The SVM moel based on featurebest(DMSFboLBP-930.69)was built in this paper.GGABE and KMCAAFS found 5 and 6 best waves separately.The union(413.259nm,428.504nm,576.064nm,411.232nm,930.69nm,416.808nm)was employed finally.After extracting texture features of every wave image,aiming at all features,models based Support Vector Machine,Back-Prognation Neural Network and Random Forrest are built.It is suggested that the classification correction of DMSFboLBP-930.69 is the highest and 91.5%.