Research And Implementation Of Atlantic Salmon Color Grading System

The fillets color of Atlantic salmon is an important indicator of its quality,taste, and growth conditions. Using the existed fillets sample collection deviceof Atlantic salmon, this article collects the fillets sample images of Atlanticsalmon in Shandong Oriental Ocean Farming Workshop of Atlantic Salmon.Firstly, this paper extracts the color features of fillets images using imageprocessing algorithms, then focuses on the methods of noise removal,background segmentation, the hole filling, the extraction of effective determinedparts, red muscle segmentation etc. Secondly, this paper reduces color featuredimensions through correlation analysis, principal component analysis(PCA)and successive projection algorithm(SPA), and establishes regression modelbased on the characteristics and color grading, respectively using the method ofBP neural network(BPNN), multiple linear regression(MLR) and least squaressupport vector machine(LSSVM). Finally, we design and develop a softwarewhich can be used for fillets color grading of Atlantic salmon.The main conclusion of this paper includes the following aspects:1. Segments the fillets and the background using OTSU, compares segmentresults under each component figure: RGB, HSV, and CIE L*a*b*.2. According to the characteristics that the boundary region of fillets isshallow and more emphasis on the middle area by sensory evaluation, wepropose an extraction method of effective determined parts. Then, consideringthe obvious white striped fillets, we segment the red muscles using K-meansalgorithm. Finally, extract nine components of effective determined parts underspace RGB, HSV, and CIE L*a*b*, and calculate their mean as a feature.3. Reduces the color feature dimensions respectively using correlationanalysis, principal component analysis (PCA), and further optimizes featureselection using successive projections algorithm (SPA). BPNN, MLR andLSSVM were used to establish regression model of fillets features and colorgrade. The results showed that, SPA has excellent feature selection effect, whenthe prediction accuracy was one level and two levels, the MLR prediction modelwe established had the correct rates of fillet color79%and97%respectively.4. Develops an Emgu CV–based fillets color grading software of Atlanticsalmon, provides alternative method of fillets image pre-processing for users. Atthe same time, users can update the parameters of this prediction model based onthe standard sample data obtained.At the end, the author summarizes the contribution of this study, andoutlook the problems existed and the further improvement methods.