The 130-page thesis of Dr. Zhejiang University teaches you to use artificial intelligence to pick watermelons.

  Artificial intelligence

If you want to ask what fruit goes best with summer, the answer must be watermelon. As a major producer and consumer of watermelons, China will consume 70% of the world’s watermelons with 20% of the world’s population in 2018, with a per capita weight of 100 kg. How to choose the sweetest melon has also become a matter of great concern to the “melon eaters”. A doctor from Zhejiang University used machine learning to judge the maturity of watermelons, which is very grounded.

Big data show that this summer is the hottest summer in 140 years. “Hot” has become a global hot word. We have experienced the hottest June, just sent off the hottest July, and are entering the hottest August …

In front of the bitter heat, holding half a piece of cold watermelon to eat is undoubtedly the best choice to relieve the heat. Eating, spoon digging, making into sand ice, stir-frying and baking are all delicious …


Significance of Watermelon to Chinese

In the world, I’m afraid there is no one who likes eating, can eat and can eat watermelon more than the Chinese.

Since watermelon was introduced into China more than 1,000 years ago during the Southern and Northern Dynasties, it has slowly begun to occupy the dominant position of summer fruits.

Historians have verified that watermelons were planted in Beijing more than 1,000 years ago, and the earliest “eating watermelons” scene of the ancients was found on the murals of Liao tombs in inner Mongolia, which were built between 1026 and 1027 a.d.


  • [Watermelon’s appearance hasn’t changed much in more than 1000 years]

When it was first introduced into China, watermelons were still reserved for aristocrats. It was not until 1143 AD that Hong Hao, an official of the Southern Song Dynasty, returned from a diplomatic mission to the Kingdom of Jin and brought back the watermelon seeds planted by the Jin people. Since then, watermelons have been found in the south of the Yangtze River and people can eat watermelons.


  • [After Hong Hao returned to the Southern Song Dynasty, he compiled his information into a book]

In Song Mo Ji Wen written by Hong Hao, there is a description of watermelons: “Watermelons are shaped like tablets with a round shape and a verdant color. They turn yellow when they are old. The cup-shaped melon tastes sweet and crisp, with juice in it, especially cold. To carry to return, now forbidden garden township constrained all have. 」

Since then, watermelons have become popular in the southern song dynasty. the words “eat watermelons everywhere in the past years”, “yellow watermelons with vines like weave” and “drunk picking watermelons to break” all indicate that watermelons have become more and more popular.


  • [In Enshi, Hubei Province, a “watermelon monument” was found in Xianchun of the Northern Song Dynasty. The inscription recorded four kinds of watermelons planted there]

Watermelon has become a part of Chinese culture. Many idioms are related to watermelons: Lao Wang sells watermelons, and when the watermelons are ripe, they fall off when they are ripe, and when they follow the vine, they touch the watermelons …

In 2004, the “China Watermelon Museum” was officially opened in Panggezhuang Town, Daxing District, Beijing. Its appearance and effect theme is “Flying Watermelon”, implying that Chinese watermelons fly out of the country. This is the first museum in China with the theme of “watermelon”. The museum shows people watermelon history, watermelon planting, watermelon culture, science and technology, etc.


  • [Watermelon-related photos, cartoons and models are everywhere in the China Watermelon Museum]

According to data released by FAO, China has become the world’s largest watermelon producer and consumer. In 2018, China digested more than 70 million tons of watermelons, or 100 kg per capita.

Therefore, in front of watermelons, we can all proudly say: I have finally realized the freedom of fruit!

The Universal Picking Melon Solution: Knock on it

However, eating melons is easier than picking melons. Facing a pile of watermelons with similar looks, how can we choose the best and sweetest watermelons? The global approach is to knock on it.


  • [Watermelon Watermelon Tell me, are you the sweetest melon]

However, in fact, many “melon eaters” finally knocked around and couldn’t hear watermelon’s answer.

As a result, a doctor from Zhejiang University wrote a 130-page paper entitled “Research on Watermelon Maturity and Acoustic Detection Technology and Device for Internal Hollow” (address:, tell us how to use artificial intelligence technology to distinguish the sweetest watermelon from the sound of knocking on it.


  • [In addition to this doctor, many doctors and masters are also doing research in this field]

Of course, his original intention in doing this research was to increase the income of watermelon farmers and expand watermelon exports. By the way, it will also benefit our “melon-eating masses.”

How to Judge Watermelon Maturity by Traditional Methods

China ranks first in watermelon production and planting area in the world, but its export volume is less than 1% of the world’s output, not only because we consume much of it ourselves, but also because the post-partum detection and treatment methods of watermelon are backward, resulting in uneven quality and low commercialization.

Whether watermelons are harvested at the right time or not has a great influence on the quality of watermelons. Over-ripe or under-ripe harvesting will affect the quality and taste of watermelons. In our country, the harvest timing mainly depends on the subjective experience of melon farmers, who often judge the maturity of watermelon by growth cycle, climate (mainly air temperature) and observation of melon skin color and texture.

Of course, there is also a skill, which is mentioned above, tapping on the melon peel to listen to the sound. This is currently the most common method used by melon farmers.


  • [Maturity of watermelon from left to right are: immature, mature, over-ripe]

Generally speaking, the sound of unripe watermelons is crisper, and when they are beaten, they will make a “thump-thump” sound. The sound of mature watermelons is lower, and their sound is “thump-thump”. Over-ripe watermelons will make a “thump-thump” sound. It takes time and effort to test one by one manually and relies on experience. The accuracy cannot be guaranteed.

Nowadays, computer, image processing, sensors and other technologies are developing rapidly. Applying these technologies to the quality inspection of agricultural products can greatly improve the inspection efficiency and accuracy.

Judging by machine learning is accurate and efficient.

Through research, the author found that the changes of watermelon maturity are generally closely related to the changes of internal components (such as sugar content) and structure (such as larger volume and hollow inside), and these changes will also lead to changes in acoustic characteristic parameters. Therefore, he thinks that it is more appropriate to use acoustic nondestructive testing methods and different modeling methods and techniques to judge watermelon maturity.

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 came to the conclusion that LS-SVM method performed best in watermelon maturity classification modeling with a prediction accuracy of 73.6%.


  • [Physical Map of Acoustic Detection Device]

On the basis of optimizing the structure and material of the acoustic detection device, and after denoising the knock signal, the author has carried out two kinds of tests, namely maturity classification test and hollow judgment test.

Kirin watermelons were selected as samples in the experiment. Five groups of watermelons farmers from Qiancun Village, Wu Shan, Cangqian Town, Yuhang District, Hangzhou were picked in batches in greenhouse and immediately moved to the laboratory after picking.

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

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

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

Four common supervised machine learning algorithms and pattern recognition algorithms are adopted in the experiment, namely linear discriminant analysis (LDA), K- minimum neighbor method (KNN), BP neural network technology (ANN) and least squares-support vector machine (LS-SVM). three watermelon varieties, immature, mature and over-ripe, are classified by acoustic characteristics, and hollow watermelon is also identified.


  • [In the hollow melon sample, the position and shape of the hollow volume are different]

In order to eliminate the adverse effects caused by peak splitting in the sound spectrum of knocking watermelons, the author defines the first-order moment index MI1 and the second-order moment index MI2. Because it is determined in the watermelon puncture test (to judge the firmness of pulp) that these two parameters are highly correlated with the maturity of watermelons, they are used as acoustic characteristic parameters for watermelon maturity classification.

Optimal melon identification algorithm: LS-SVM

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


  • [LS-SVM Classification Results of Three Watermelons with Different Maturity]

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

The author concludes that LS-SVM uses linear equations to simplify the quadratic programming problem of SVM, greatly improves the classification efficiency, and introduces penalty factors to minimize structural risks. The method is more suitable for processing the classification of small sample set imbalanced sample set imbalanced watermelon sample set.

LDA algorithm is suitable for linear classification problems. When KNN algorithm classifies unbalanced sample sets, it is easy to bias the position samples to a large number of sample classes, resulting in larger classification errors. Neural network classifier has the disadvantages of over-fitting and easy falling into local extremum.

In addition, the author also uses the above four kinds of methods and the empirical threshold method of energy ratio, taking MI1, MI2 and energy ratio Er as characteristic parameters, to distinguish hollow melons from unbalanced samples.

The author uses Fβ score as the evaluation index of the classifier. Its physical meaning is to combine the accuracy rate and recall rate into a score by giving them different weights to comprehensively evaluate the accuracy of the classifier in classifying the sample population and certain samples.


  • [definition of fβ]

In the definition formula, TP is the number of samples correctly divided into hollow melons, TN is the number of samples correctly divided into good melons, p is the actual number of hollow melons, and n is the actual number of good melons. In this paper β = 2.

Among these discrimination methods, LS-SVM still has the best classification effect, and its Fβ scores of modeling set and prediction set are 88.1% and 74.7% respectively.

It seems that Dr. Zhejiang University has helped us calculate clearly how to choose the sweetest melon.

Eat melons, we are the most scientific

With the accumulation of thousands of years, watermelon has become more and more a part of Chinese life. The so-called “people who have not eaten watermelons in summer cannot speak of life. 」

Nowadays, the very popular online word “melon eaters” also fully shows our love for watermelon. Originally, “melon” in “melon eaters” meant melon seeds, but as watermelon became popular, the “melon” gradually evolved into “watermelon”.


For students in the field of artificial intelligence, the most familiar “melon” must be the watermelon book “Machine Learning” written by Professor Zhou Zhihua of Nantah.


  • [This introductory book on machine learning starts with picking watermelons and explains various terms and problems with watermelons everywhere]

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


  • [Japan Shibuya Company’s Watermelon Picking Equipment, next time you can take it to buy melons]