A 130-page Paper by a Zhejiang University Doctor Teaches You How to Pick Watermelons Using Artificial Intelligence

If you ask what fruit goes best with summer, the answer must be watermelon. As a major producer and consumer of watermelons, China consumed 70% of watermelons in 2018 with a population of 20%, or 100 catties per person. How to choose the sweetest watermelon has also become a matter of great concern to the "watermelon-eating crowd". A doctor from Zhejiang University used machine learning to judge the maturity of watermelons, which is very down-to-earth.

Big data shows that this summer is the hottest in 140 years. "Heat" has become a global hot word. We have experienced the hottest June, just bid farewell to the hottest July, and are entering the hottest August...

In the face of the unbearable heat, holding half a piece of coldwatermelonEating it with relish is undoubtedly the best way to cool down. You can eat it by gnawing, digging it with a spoon, blending it into smoothies, frying it, grilling it, it tastes good in any way...

What watermelon means to the Chinese

In the world, there is probably no group of people who love to eat watermelon more, eat more and know how to eat watermelon better than the Chinese.

sinceMore than 1,000 years agoDuring the Northern and Southern Dynasties, watermelon was introduced to China, and it gradually began to rise to the status of the dominant summer fruit.

According to historians, more than 1,000 years agoBeijing areaWatermelon has been planted in China for a long time. The earliest scene of people eating watermelon was found on a mural in a Liao Dynasty tomb in Inner Mongolia.Between 1026 and 1027 AD.

When watermelon was first introduced to China, it was only enjoyed by the nobility.Hong Hao, an official of the Southern Song DynastyWhen he returned from his mission to the Kingdom of Jin, he brought back watermelon seeds grown by the Jin people.From then on, watermelons were available in Jiangnan area and common people could eat them.

In Songmo Jiwen written by Hong Hao, there is a description of watermelon: "Watermelon is round and shaped like a plaque, and its color is very green, but it turns yellow after a year. Its stem is similar to that of a melon.**It tastes sweet and crisp, juicy and especially cold.**I brought it home, and now it is in all the forbidden gardens and village parks."

After that, watermelon began to appear frequently in the writings of Southern Song literati. Words such as "watermelon has been eaten everywhere in recent years", "where the watermelon is yellow, the vines are like weaving", and "picking up watermelons while drunk and splitting them" all indicate that watermelon has become popular.Becoming more and more popular.

Watermelon has become a part of Chinese culture. Many idioms are related to watermelon: Lao Wang sells watermelons, the melons are ripe, the melons are ripe and fall off the vine, follow the vines to find the melons...

**In 2004,**Located in Panggezhuang Town, Daxing District, Beijing **"China Watermelon Museum",**The museum is officially opened. Its appearance theme is "Flying Watermelon", which means that Chinese watermelons are flying abroad. This is the first museum in China with "watermelon" as its theme. The museum shows people the history of watermelon, watermelon planting, watermelon culture and technology.

according toFood and Agriculture Organization of the United Nations (FAO) According to published data, my country has become the world's largest producer and consumer of watermelons.2018my country has digested more than70 million tonsWatermelon, that is, per capita consumption100 jin.

So, in front of the watermelon, we can all proudly say: I have finally achieved fruit freedom!

The universal method of picking melons: knock it

**However, it is easy to eat melons but difficult to choose them.**When faced with a bunch of similar-looking watermelons, how can you choose the best and sweetest one? The universal method is: knock it.

But, in fact, many "melon-eating people" knocked and knocked but couldn't hear any answer from the watermelon.

So, aPhD from Zhejiang UniversityWrote an article130 pagesPapers***"Study on Acoustic Detection Technology and Device for Maturity and Internal Hollowness of Watermelon"******Paper address:***http://t.cn/AiYcAYei******）, tells us how to use artificial intelligence technology to identify the sweetest watermelon from the sound of knocking on it.

Of course, his original intention of doing this research was to increase the income of watermelon farmers and expand the export of watermelons, and also to benefit our "watermelon-eating masses".

How to judge the ripeness of watermelon by traditional method

my country's watermelon production and planting area rank first in the world, but the export volume is less than 1% of the production, not only because we consume more ourselves, but also becauseThe means of post-harvest inspection and processing of watermelons are backward, resulting in uneven quality and a low level of commercialization.

Whether watermelons are harvested at the right time has a great impact on their quality. Picking them too ripe or too unripe will affect their quality and taste.**The timing of harvesting mainly depends on the subjective experience of melon farmers.**They often judge the ripeness of watermelons by their growth cycle, climate (mainly temperature) and by observing the color and texture of the rind.

Of course, there is another trick, which is the one mentioned above, which is to knock on the melon skin and listen to the sound. This is the most commonly used method by melon farmers.

Generally speaking,**Unripe watermelons have a crisper sound.**When you knock it, it makes a "dong dong dong" sound.**The sound of a ripe melon is relatively low and turbid.**The sound is "bang bang bang", while overripe watermelons make a "pop pop pop" sound. Manual inspection one by one is time-consuming, labor-intensive and experience-dependent, and the accuracy cannot be guaranteed.

Nowadays, technologies such as computers, image processing, and sensors are developing rapidly. Widely using these technologies in agricultural product quality testing can greatly improve testing efficiency and accuracy.

Use machine learning to make judgments, accurate and efficient

Through research, the authors found that changes in watermelon maturity are generally closely related to changes in internal composition (such as sugar content) and structure (such as increased volume and mass, and internal hollowness), and these changes can also lead toAcoustic characteristic parameterschanges, so he believes that using**Acoustic nondestructive testing methods,**It is more appropriate to judge the ripeness of watermelon through different modeling methods and techniques.

Moreover, compared with other detection methods, such as laser, nuclear magnetic resonance and other technologies, acoustic feature detection has the advantages of low price, high detection efficiency and high accuracy. After the experiment, the author concluded thatLS-SVM methodThe best performance in watermelon maturity classification modeling,The prediction accuracy is 73.6%.

Based on the optimization of the structure and materials of the acoustic detection device and after denoising the knocking signal, the authors conducted two experiments:Maturity classification test and hollow determination test.

Qilin watermelon was selected as the sample in the experiment. The watermelon was picked in batches by the melon farmers from the fifth group of Wushanqian Village, Cangqian Town, Yuhang District, Hangzhou City in the greenhouse melon shed, and then moved to the laboratory immediately after picking.

In the maturity classification experiment, 147 non-hollow melons were selected and randomly divided into two groups: 75 in the modeling set and 72 in the prediction set.

In the hollow determination test, 190 samples (including hollow melons) were selected and randomly divided into two groups, 97 in the modeling set and 93 in the prediction set.

The modeling set is used to establish the classification or hollow judgment model of the samples, and the prediction set is used to test the performance of the model.

The experiment used four commonSupervised Machine Learning Algorithmsand pattern recognition algorithms, respectively.**Linear discriminant analysis (LDA), K-least nearest neighbor (KNN), BP neural network (ANN) and least squares-support vector machine (LS-SVM),**The acoustic characteristics were used to classify three types of watermelons: unripe, ripe and overripe. In addition, hollow watermelons were also identified.

In order to eliminate the adverse effects of peak splitting in the spectrum of the sound of tapping watermelon, the authors defined**one****The first-order moment index MI1 and the second-order moment index MI2,**Because in the watermelon puncture test (to determine the firmness of the flesh), these two parameters are highly correlated with the maturity of the watermelon.It is used as the acoustic characteristic parameter for watermelon maturity classification.

The most suitable melon identification algorithm: LS-SVM

The experimental results show that the accuracy of the modeling set and prediction set of the LS-SVM algorithm are 76% and 73.6% respectively, which are higher than the classification results of the other three classifiers.

In addition, the accuracy of the neural network modeling set and prediction set are 73.3% and 66.6% respectively, second only to LS-SVM.

The authors conclude:LS-SVM uses a linear system of equationsThe quadratic programming problem of SVM is simplified, and the classification efficiency is greatly improved.Introducing a penalty factorMinimize structural risk. This method is more suitable for processing small sample sets, unbalanced sample sets, and unbalanced watermelon sample sets;

andLDA The algorithm is suitable for linear classification problems;KNN When the algorithm classifies an unbalanced sample set, it is easy to make the position samples biased towards the sample class with a large number, resulting in a larger classification error;Neural NetworksThe classifier has the disadvantages of overfitting and being prone to falling into local extreme values.

In addition, the authors also used the above four methods and the energy ratio empirical threshold method to calculate the energy ratio based on MI1, MI2 and energy ratio E.r is the characteristic parameter, and the imbalanced samples areIdentification of hollow melons.

The author uses Fβ The score is used as an evaluation indicator of the classifier. Its physical meaning is to give different weights to the precision and recall, combine the two into one score, and comprehensively judge the accuracy of the classifier in classifying the sample population and a certain type of sample.

In the definition formula, TP is the number of samples correctly classified as hollow melons, TN is the number of samples correctly classified as good melons, P is the actual number of hollow melon samples, and N is the actual number of good melon samples. In this paper, β = 2.

Among these identification methods,**LS-SVM has the best classification effect.**The F of its modeling set and prediction setβThe scores are 88.1% and 74.7% respectively.

It seems that a PhD from Zhejiang University has already figured out for us how to pick the sweetest melon.

Eat melon, we are the most scientific

After thousands of years of accumulation, watermelon has become more and more a part of Chinese people's life. As the saying goes, "If you have not eaten watermelon in summer, you are not qualified to talk about life."

Nowadays, very popular network buzzwords **"Melon-eating crowd"**It also fully demonstrates our love for watermelon. Originally, the "melon" in "melon-eating crowd" referred to melon seeds, but as watermelon became popular among the people, the "melon" gradually evolved into "watermelon".

For students in the field of artificial intelligence, the most familiar "melon" must be the one written by Professor Zhou Zhihua of Nanjing University. Machine Learning Watermelon Book.

It seems that Professor Zhou Zhihua must also like eating watermelon very much.

-- over--

Click to read the original article