Design and Data Analysis of Muskmelon Sugar Content Detection System

Sugar content of muskmelon is an important indicator to reflect the maturity of muskmelon, and also an important factor to determine the commercial grade of muskmelon. Aiming at the disadvantages of traditional muskmelon detection methods, the feasibility of applying acoustic characteristics combined with machine learning to nondestructive detection and grading of muskmelon was investigated. The muskmelon acoustic detection system was designed, the time domain signals of different batches of samples were collected. After the time domain signal is normalized, the frequency domain signal is obtained by fast Fourier transform method, and the data signal is pretreated to remove the tendency. The principal components of frequency domain signals are extracted by principal component analysis (PCA), of which the cumulative variance contribution rate of the first 3 principal components is 95.22%, and the 1 and 2 principal components are separable for samples of different levels. Four different machine learning algorithms were applied to establish a muskmelon full variable classification model, and the accuracy of the verification set classification was more than 67%. The stable competitive adaptive weighting algorithm is signed to extract the feature variables, reducing the number of variables by about 84.27%, the performance of the classification model established by using the optimized feature variables is improved. The support vector machine model (SVM) has the highest accuracy of verification set (95.74%), score (96.49%) and Kappa coefficient (93.71%). The results show that it is feasible to combine acoustic characteristics with machine learning method for nondestructive testing and grading of muskmelon.