Research Tomato Quality Detection Based On Computer Vision

Tomato is one of the most popular vegetables all around the world.It has rich nutritional value,and its sweet and sour taste is another reason for its popularity among consumers.The quality of tomato is an important standard for assessing its value.Visual description of tomato value is an important reference for consumers when buying tomatoes.The external quality of tomatoes is the main factor affecting consumers’ willingness to purchase it,and the changes of its external quality can lead to physiological reactions,which will eventually affect the internal quality of tomato.This essay took the tomato as its research subject,using computer technology to detect and classify tomato quality.Meanwhile,it developed a tomato quality detection system,which laid the foundation for automatic grading and evaluation of tomato quality.The main research contents of the thesis are as follows.(1)Explore image preprocessing methods.In order to obtain high quality of images,this study designed and build a visible computer vision system to minimize the impact of the external environment on image quality.Meanwhile,for the purpose of eliminating the original noises and improving the images quality,based on the analysis of the advantages and disadvantages of different image filtering methods,this research finally chose median filtering as the preprocessing method to eliminate the original graphic noise image.(2)Feature extraction method for tomato image.The characteristics of tomato itself were combined to analyze the characteristics of tomato images;the color and shape features of tomato images were extracted from two perspectives of maturity(color)and shape.RGB and HIS model were used to extract the color features of tomatoes,and based on Hu invariant moment,tomatoes’ shapes were extracted.The experiments have shown that the extracted features can effectively reflect the color and shape quality characteristics of tomato.(3)Establish a tomato external quality inspection model.The Least Squares Support Vector Machine(LSSVM)model uses least squares as the loss function,which can effectively reduce the computational complexity and improve the computational efficiency.It has obvious advantages in the processing of small and nonlinear data,so the research extracted tomato characteristics(12 color features and 7 shape features)were taken as input.The tomato shape detection model was established by using the least squares support vector machine method.Because GA genetic algorithm has superior performance in global optimization,this study uses genetic algorithm(GA)to optimize the parameters of the model,reduce the root mean square error of the model as much as possible,and improve the accuracy and robustness of the model.When the penalty coefficient C was 46.64 and the nuclear parameters was 23.04,the error of root mean square minimized to 0.79.The recognition rate of GA-LSSVM system on the training set was 86.88%,and its recognition rate on test set was 83.38%.(4)Establishment of tomato sugar detection model based on convolutional neural network.Firstly,number the tomatoes and get the image,then measure the sugar level of the numbered tomatoes and record the data.Secondly,the data was expanded in a ratio of 1:15.This paper also introduces the basic principle of convolutional neural network,adopts Alex-Net structure,sets model initialization parameters,and establishes tomato sugar content prediction model based on convolutional neural network.(5)Based on Django framework and Open CV,an intelligent computer vision tomato grading system was designed and developed.The system’s functions consists of registration and login,tomato images preprocessing,tomato features extraction,and tomato quality detection.The system itself has strong portability,so it is easy to deploy online.With its simple and clear operation,it is very user-friendly.