:::

【2020 Application Example】 Automatic fruit screening system: A solution that uses neural networks, AI, and automation to improve fruit screening efficiency by 10 times, increase output value by NT$1.7 billion, and significantly improve quality with 93% accuracy

Taiwan is located in the subtropics and has a diverse geographic environment that is very suitable for growing fruit. Bananas and pineapples were once extremely popular export commodities that we are proud of. However, farmers in consuming countries gradually obtained the excellent seeds of Taiwan’s fruits, and were able to grow the same quality fruit but at a more affordable price, causing our fruit exports to face a major crisis! At present, although Taiwan's fruits such as mango and guava still have certain competitive advantages, if they fail to make further progress compared with other countries, they will still encounter the same problem over time and cannot be ignored! Fruit quality and brand value are the only ways for Taiwan's fruit industry to remain competitive internationally.

Fruit screening is the main link in fruit production and marketing that determines quality. Currently, the industry is highly dependent on aging rural manpower, resulting in rising fruit screening costs due to labor shortage and making it extremely difficult to maintain stable yield. Therefore, the automation of fruit screening work has become a very important and urgent issue. Professor Chi-Chun Lee at the Department of Electrical Engineering of National Tsing Hua University led a team to develop an automatic fruit screening system that combines cameras, conveyor belts, and AI. The system currently has an accuracy reaching 93%. One production season can increase the output value of mango by NT$1.7 billion. With the gradual development of the AI system, the accuracy is expected to improve in the future, and the same system can also be applied to other fruits, further promoting traceable fruit and driving the technological upgrading of Taiwan's fruit industry.

Fruit screening relies heavily on scarce manpower, and the aging of the rural population makes the situation even worse

Professor Chi-Chun Lee learned about the fruit industry’s dilemma from his classmate Yu (alias), who had studied together in the United States. Yu is the young second-generation successor of one of Taiwan's largest fruit import and export companies. According to Yu's observations in the industry over numerous years, Taiwan's fruit production and export usually generated good profits at first, but after fruit farmers in the consuming countries obtained the seeds, they will often attempt to grow the fruit locally to reduce costs and obtain greater profits. If Taiwanese fruits cannot surpass the products of fruit farmers in consuming countries in terms of quality or brand value, they will be eliminated because competitors' costs are indeed lower.

Fruit screening is used to divide fruits according to quality. If they cannot pass the minimum specification, they will be discarded as waste products. In practice, the work of screening fruits will be carried out by farmers' goods yards and distributor' packaging yards respectively. However, if it is not properly handled by the collection freight yards and the packaging yards do not do a good job in sampling in the early stage, it will result in a loss for distributors and cause 30% of A/A+ grade fruits to be eliminated.

This job relies heavily on experienced fruit screeners. More experienced fruit screeners can not only control the quality and reduce the chance of fruit damage in the fruit screening process, but also have the ability to pick out about 10% more A+ grade fruits, which adds great value. What worries the industry is that experienced fruit screeners are gradually decreasing due to the aging population in rural areas, making them a very rare resource. Such rare human resources are often in high demand during busy farming periods. Farmers or distributors who fail to hire experienced fruit screeners have to settle for less experienced one, taking on the risk of additional losses and paying greater costs. The most unfortunate situation suffering a loss of 30% mentioned above.

▲ Fruit screening is an important process in the later stages of fruit production when packaging and selling. Failure to properly control quality will result in huge losses.

AI is very suitable for assisting in fruit screening, but it is difficult to obtain data sets

After understanding Yu's difficulties, Professor Lee found that this was a problem that could be solved using AI - although fruit screening relies heavily on experienced fruit screeners, it is a highly repetitive task. Handling repetitive tasks with a large amount of data has always been a strength of AI.

However, the first problem appeared even before research and development work started: Which fruit do we start with?

First of all, a suitable fruit must reach a certain export volume, and the fruit must still have considerable room for growth. For some fruits that lack international competitiveness, such as bananas and pineapples, companies no longer have the ability to invest more funds to purchase equipment, let alone sponsor R&D or assist the R&D team in experiments.

When you have an idea, you need to pick up the pace and put it into practice as soon as possible! Therefore, Irwin mango, which still has a certain advantage in terms of scale, was selected as the first experimental subject of the automatic fruit screening system.

The first step after harvesting mangoes is to screen the fruits for the first time at the goods yard. After the fruits are screened, they are sent to the packaging yard for fumigation and disinfection, and preparation for sale or loaded into containers for export. However, exporters with a deeper understanding of the target market will have stricter quality requirements and will often screen the fruit again to ensure the quality of the fruit before fumigation at the packaging site. Since employees at the goods yard are paid based on the number of mangoes screened rather than on the quality of the mangoes, they focus on quantity when working. As a result, to ensure the quality of the selected fruits, the subsequent packaging factory has to screen the fruit again, increasing labor. The solution seems simple and clear - A camera, machine conveyor belts for grading and sorting, and an AI that can distinguish the quality of mangoes from their appearance are all that are needed to achieve automatic fruit screening. However, the hard part is how can AI distinguish the quality of mangoes? That’s right, you must start by establishing a training data set! In order to create the data set, Professor Lee's team established a website that allows anyone to upload photos of mangoes and rate them. Once the data sets are refined, they can be used to train AI.

▲ The fruit screening machine developed by Professor Lee's team uses AI image recognition to select the best looking mangoes.

The accuracy of the trained AI reaches 93%, which can increase the output value by NT$1.7 billion in one season.

In 2019, the assistance of the Industrial Development Bureau (now the Industrial Development Administration of the Ministry of Economic Affairs) and AI HUB accelerated the verification of the technology.

Professor Lee's team accumulated 100,000 entries of data during the 2-month empirical period, and the accuracy of the trained AI reached 93%! This is far higher than the manual screening accuracy of 70%, resulting in a clear difference in quality. In terms of export value, the output value of mango is expected to be increased by NT$1.7 billion in one season! It can also reduce labor costs by NT$1.866 million and avoid the seasonal labor shortage problem mentioned above.

In addition, since it is no longer necessary to screen the fruit once at the goods yard and packaging yard each, it also reduces losses caused by human error in the fruit screening process. When the technology becomes more mature, the same system can be applied to other fruits exported by Taiwan, such as wax apple and guava, in the future, taking Taiwan's fruit industry to the next level.

Since it is AI, accuracy can be improved through continuous training, and continuous adjustment of algorithms and cooperation with equipment manufacturers can significantly improve production capacity. In addition, Professor Lee is also organizing the AI Cup competition with the sponsorship of manufacturers and the government, allowing more teams to use the same data set to continue to develop the algorithm, in hopes of facilitating further cooperation with companies that are interested.

Irwin mango grade identification system on AI HUB

Professor Lee's team hopes to use the power of AI to achieve complete traceability of fruits from production to packaging and transportation, thereby increasing the brand value of Taiwan's fruits! Besides hoping to allow Taiwan's fruits to seize a place in the fiercely competitive foreign markets, with high-quality supply, Taiwan's fruits can also shine internationally and become the pride of Taiwan.

▲ Taiwan's fruits still have certain competitive advantages in the international market, but they also face competitive pressure from fruit farmers in consuming countries as they are exported.


▲ Easily save NT$1.866 million per mango season and significantly improve quality.

 

 

 

 

 

 

Recommend Cases

【導入案例】巨量遙測空間數據AI分析雲端服務平台 使衛星遙測影像順利落地應用
AI Analysis Cloud Service Platform for Remote Sensing Big Data Enables the Smooth Application of Satellite Remote Sensing Images

Although satellite remote sensing images can make all surface objects visible, it still requires a lot of time and manpower to be truly applied to the industry In order to effectively solve the problems that customers face in digesting huge amounts of image data and eliminate technical obstacles for cross-domain users to process satellite remote sensing images, ThinkTron has developed an "AI Analysis Cloud Service Platform for Remote Sensing Big Data" as a new beginning for cross-domain AI applications for spatial information In recent years, in response to the impact of industrial globalization, Taiwan's agriculture has continued to transition towards technology-based and higher quality, improving the yield and quality of crops by solving problems, such as microclimate impacts and pest and disease control The demand of agriculture on images has expanded endlessly to accurately grasp the growing environment of crops In the early years when UAVs unmanned aerial vehicles were not yet popular, manual field surveys were the most basic but most labor-intensive work With the emergence of UAV drones, aerial photography operations might not be difficult, but the range that can be photographed is limited Furthermore, surveying expertise is required to accurately capture spatial information At this time, the use of satellite remote sensing data may break away from the past imagination of using image data Taiwan Space Agency TASA ODC data warehouse services In the past ten years, with the breakthrough of modern satellite remote sensing application technology, Digital Earth has become a new trend in global data collection Countries have developed data cube image storage technology, and the development of smart agriculture has become one of the largest image users Determining planting distribution is the first step in understanding crop yields Free satellite remote sensing images, powerful data warehousing support, and the team's robust image recognition technology are important supports for accelerating agricultural transformation Using satellite remote sensing image data can accelerate the development of smart agriculture However, in the past, it was difficult to extract large-area crop distribution through satellite remote sensing images, not to mention the cost If you wanted to use free information, you had to visit all websites of international space agencies, look through the wide variety of satellite specifications, and carefully evaluate the sensor specifications, image resolution, and revisit cycle After finding suitable images, you had to look at them one by one to filter the ones you need Next is downloading dozens of images that are often several hundreds of Megabytes MB each, which might exceed the capacity of your computer Also, after overcoming image access and preparing data, you must then start to confirm the downloaded image products and which bands you want, because the image you see is not just an image file jpg or png, but rather complex multi-spectral information, attribute fields and coordinate information It takes a lot of effort just to confirm the correct information Facing GIS software packages with complex functions is the start of another trouble The complex image pre-processing process and the inflexible machine learning package greatly reduce the efficiency of analyzing data After finally getting the results of crop identification, you might find that the best time for using map information may have already passed The above-mentioned complex and time-consuming satellite image processing problems are precisely the pain points of the market ThinkTron expanded from traditional machine learning to modern deep learning applications, and developed an "AI Analysis Cloud Service Platform for Remote Sensing Big Data" under the GeoAI framework, breaking through the constraints of details in the spatial information for customers Differences between the process before and after introducing the AI analysis cloud service platform ThinkTron said that Taiwan's ODC Open Data Cube system has been completed and began providing services after years of efforts from the Taiwan Space Agency TASA, formally becoming aligned with international trends The powerful warehousing technology allows users to easily capture and use image data of a specific time and spatial range according to their needs The warehouse stores multiple satellite image resources from international space agencies, including the ESA's Sentinel-1 one image every 6 days, Sentinel-2 one image every 6 days, USGS's Landsat-7 one image every 16 days, Landsat-8 one image every 16 days, and the domestic Formosat-2 one image every day and Formosat-5 one image every 2 days ThinkTron develops satellite image recognition tools based on Python Breaking free from the limitations of GIS Geographic Information System software packages, ThinkTron integrated GDAL Geospatial Data Abstraction Library based on Python, and considered computing efficiency and parallel processing when developing all tools required for satellite image processing and image recognition modeling, including coordinate system and data format conversion, grid and vector data interaction, and data intra-difference and normalization All of the tools are designed with AI applications in mind, and some commonly used tools are packaged into an open source package under the name TronGisPy to benefit the technical community ThinkTron utilized the team's understanding of satellite remote sensing images and the collected tagged data crop distribution maps to preset the image recognition modeling process, the required training data specifications, and dataset definitions This is imported into the machine learning LightGBM or deep learning CNN framework that was completed in advance, and the entire training process to be performed in the Web GIS interface, providing users with partial flexibility to freely filter images, confirm spatial and temporal ranges, select models, and adjust hyperparameters In addition to the operation of training models, it also provides historical models to output identification results, and finally displays the identification results of crop distribution on the Web GIS map In fact, agriculture is not the only industry that needs satellite remote sensing applications AI applications of spatial information have also appeared in various fields as companies in different industries aim to enhance their global competitiveness For example, surveying and mapping companies that have a large amount of map data can use the AI analysis cloud service platform to store map data while also accelerating the efficiency of digital mapping Under the severe global climate change and the risk of strong earthquakes, there is a wide variety industrial insurance, agricultural insurance, financial insurance, or disaster insurance are all inseparable from spatial information The use of remote sensing image recognition to understand insurance targets has long been an international trend AI Analysis Cloud Service Architecture for Remote Sensing Big Data

【導入案例】海量數位工程AOI機器智能手臂檢測系統 大幅提高瑕疵檢測精準度
Massive Digital Engineering AOI Intelligent Robotic Arm Inspection System Significantly Improves Defect Detection Accuracy

Taiwan is known as a manufacturing powerhouse, yet quality defect detection has always been a chronic sore point in production lines While AOI equipment is available to assist, most use fixed machinery which are limited by angles, resulting in less precise diagnostics and high false positive rates Massive Digital Engineering introduced an AOI intelligent robotic arm detection system that effectively reduces false positives and increases the accuracy of defect detection Generally, the yield rate of products affects the costs for enterprises and the return rate for customers The quality defect detection process in the manufacturing industry often necessitates a substantial amount of quality inspection labor Although there is AOI equipment to assist, these tools are mostly fixed detection machines Fixed cameras are easily limited by angles, resulting in less precise diagnostics and high false positive rates Thus, personnel need to re-screen and inspect afterwards, often manually visual inspection misses defects on average about 5, and can be as high as 20 Three major pain points in manufacturing quality detection Robotic Arm AOI with dynamic multi-angle inspection helps to solve these issues According to the practical understanding by Massive Digital Engineering, there are three main pain points in detecting product quality within the manufacturing industry Pain point one, manual inspection of product quality is prone to errors Currently, the manufacturing industry largely relies on human labor to inspect product appearance, but human judgment often entails errors, such as surface scratches, color differences, solder appearance, etc The error rate in defect judgment is high, and can only be inspected at the finished product stage, often leading to whole batch rejections and high costs in labor and production Pain point two, inability to quantify and record data from quality inspections Traditional manual inspections do not maintain inspection data, which makes it difficult to assign responsibility when quality disputes occur Moreover, high-end contract manufacturing orders from overseas brands often require traceability and corresponding defect records, which traditional human inspection methods struggle to meet Pain point three, limitations of traditional AOI visual inspection systems Current manufacturing uses AOI visual inspection systems, which due to the limitations of visual software technology, employ fixed cameras, fixed lighting, and single-angle operations This method may handle flat or linear-shaped products like rectangular or square items at a single inspection point However, it is more challenging to implement for products with complex shapes eg, irregular automotive parts, requiring multi-point and multi-degree inspections Massive Digital Engineering developed an AOI intelligent robotic arm detection system, effectively improving the accuracy of defect detection To address the pain points in quality inspection in manufacturing, Massive Digital Engineering initiated the concept of developing a multi-angle, movable inspection device, starting with the combination of two representative technologies in factory automation - robotic arms and machine vision By integrating robotic arms with AOI for dynamic multi-angle AI real-time quality inspection, the limitations of fixed inspection systems are addressed, and visual inspection techniques are enhanced by leveraging artificial intelligence, further elevating the sampling of images from flat to multi-dimensional and multi-angular Selected the automotive industry as the real-world testing ground to quickly respond to customer needs The AOI intelligent robotic arm detection system, utilizing AI technology including unsupervised learning, supervised learning, and semi-supervised learning, allows operators to use unsupervised deep learning techniques to learn about good products even when initial samples are incomplete or there are no defective samples, applying it in the visual inspection of automatic welding of car trusses This can solve issues of limited angles with fixed machinery before implementation, less precise diagnostics, and high false positive rates Automotive components are high in unit price and demand a stricter defect detection accuracy In industries that have adopted AI services, the automotive manufacturing sector was chosen as the real-world testing ground Massive Digital Engineering states that the automotive industry mainly consists of related component manufacturers and components typically have a higher unit price, hence requiring more in terms of quality inspection and yield rates, and demanding stricter accuracy Therefore, the automotive sector was chosen as the area for introduction By using a robotic arm combined with AI for dynamic multi-angle AOI visual real-time quality inspection, not only can the defect quality error rate of automotive components be improved, but the fixed-point AOI optical inspection can be enhanced to meet the measurement needs of most industries and finally, establishing a third-party system platform to build an integrated monitoring system platform, enabling immediate response and action when issues arise This system allows for recording and storing important data of products leaving the factory, serving as a basis for future digital production lines and virtual production At the same time, in the event of defects, it can immediately connect to Massive's MES monitoring system, quickly responding to the relevant manufacturing decision-making department, subsequently utilizing ERP systems for project management and reviews, effectively improving production efficiency and reducing production costs Helps to reduce communication costs and aims to become an industry standard In terms of industry integration, it provides a foundational standard for data continuity among upstream and downstream businesses, reducing communication costs within the supply chain Through certification of the contract manufacturers and brand owners, there is a chance to become the industry standard configuration Through the data database established by this project, operators can further optimize their supply chain management solutions using big data analysis Data Analysis, based on data, establish forecast planning, and utilizing technology to link upstream and downstream data of the supply chain, accurately controlling product quality In the future, when interfacing with European, American, and Japanese markets, which demand highly fine-tuned orders, operators can respond and integrate the industry supply chain Supply Chain more swiftly Ultimately, through the benchmark demonstration industry's field verification, such as with the automotive component manufacturing industry used as the benchmark demonstration field, by implementing the robotic arm combined with AI for dynamic multi-angle AOI visual real-time quality inspection system project, the supply chain connection between automotive contract manufacturers and OEMs can be optimized, becoming the industry standard Further seeking more AI teams to join the cross-industry development on the field collaboration platform, driving the overall ecosystem combining AI innovation with field application Self-driving vehicle developed by Massive Digital Engineering「Translated content is generated by ChatGPT and is for reference only Translation date:2024-05-19」

【解決方案】優式AI智能割草機器人 搶攻高爾夫藍海市場
USRROBOT's AI Lawn Mowing Robot Enters the Blue Ocean of Golf Market

An AI smart lawn mowing robot, resembling a vacuum robot, shuttles back and forth on the 30-hectare golf course lawn for weeding This robot, independently developed and designed by Taiwanese, is equipped with the world's first electronic fencing positioning technology which utilizes high-precision GPS integrated with cloud AI computation to determine the most efficient mowing paths, targeting the lucrative blue ocean market of golf courses This AI lawn mowing robot was developed by USRROBOT, a Taiwanese startup established in 2019 Chao-Cheng Chen, the president of USRROBOT, once served as the executive vice president of one of the top five ODM tech companies in Taiwan, and specializes in software and hardware integration When he served as the chairman of the Service Robot Alliance, he knew that the service robot industry was bound grow rapidly due to declining birth rates and the growingly severe labor shortage New demand - The horticulture market is large and the has rigid demand "To develop the core technology of service robots, we must find rigid demand Looking at European and American countries, there is a shortage of labor, but demand for horticulture has increased, and there has been a long-term shortage of 7-10 of horticultural workers" Under this strong "rigid demand," Chao-Cheng Chen established USRROBOT, and the company's first product is the AI lawn mowing robot In terms of overseas markets, the United States is the world's largest horticulture market, accounting for 30-40 of the global output value It is estimated that there are about 1 million horticulture workers, but they have been experiencing a labor shortage of 7-10 in recent years and have not been able to improve the situation The main reasons for labor shortage are Aging population and gardening is a labor-intensive job, so young people don't want to do it Unlike in Taiwan, European and American countries attach great importance to lawn maintenance and have expressly stipulated in the law that heavy fines will be imposed for failing to mow the lawn Therefore, the AI lawn mowing robot has considerable market development potential The introduction of AI multi-device collaborative mowing sensor technology is expected to reduce the burden of staff maintaining the golf course The AI lawn mowing robot developed by USRROBOT is currently in its second generation Domestic universities and well-known art museums are using the latest model M1, and it is also being used by some world-renowned high-tech companies and well-known universities in the United States The company is currently involved in negotiations for subsequent business cooperation USRROBOT stated that the professional RTK system currently used can reduce the original GPS positioning error from tens of meters to about 2 centimeters, allowing the robot to move accurately outdoors After setting the boundaries, it can be easily operated using the app New application - Implementation in golf courses solves the problem of labor aging and shortage Chao-Cheng Chen further explained that the National Land Surveying and Mapping Center is a RTK service provider RTK provides the error reference map of the positioning point USRROBOT can access the positioning error value of a specific position through 4G Internet access The AI algorithm of USRROBOT reduces the general 10-20 m error of GPS to 2 cm After positioning, USRROBOT then uses six-axis accelerator positioning, gyroscopes, and wheel differential sensing devices for software and hardware integration Only by matching the wheel's movement pattern and the terrain can accurate mowing path planning be achieved The AI lawn mowing robot, which is 62 cm wide, 84 cm long, 46 cm high, and weighs only 25 kg, can set the mowing boundaries in the cloud It can avoid pools and sand pits through settings, using AI algorithms to automatically calculate the optimal path It is able to mow approximately 150 ping of grass in one hour The battery can be used continuously for more than 6 hours The battery life is currently the highest in the world In addition to general gardening companies, with the assistance of the AI project team of the Industrial Development Bureau, Ministry of Economic Affairs, USRROBOT's AI lawn mowing robot has been applied to golf course lawn mowing A well-known golf course located in Taiping District, Taichung City currently has a staff of 5 people who are responsible for the lawn, planting maintenance, and other landscape maintenance of the entire 30-hectare course However, the average age of staff is as high as 55 years old, and the golf course has been unable to recruit new staff members for a long time In view of the aging staff and the shortage of manpower, the golf course hopes to mitigate the impact with AI technology, and is therefore using AI multi-device collaborative mowing sensor technology, in hopes of reducing the burden of staff maintaining the golf course New challenges - Expert systems are needed to overcome difficulties with different grass species "This AI lawn mowing robot has low noise, low pollution, low labor costs, and is waterproof and anti-theft In the lawn mowing process, it can identify and avoid obstacles through ultrasonic sensors while maintaining mowing quality, maintaining aesthetic and consistent grass length" Chao-Cheng Chen went on to say that the most important part about golf courses is that the grass pattern should be beautiful and free from diseases and pests Based on the site survey, golf courses are mainly divided into three major areas green, fairway and rough There is no problem using the current mowing robot to mow the rough area, and it can overcome slopes within 20 degreesThe short grass in the fairway area may only be two centimeters long, and the grass types are also different, so the cutterhead design needs to be modifiedAs for the grass in the green area, the grass must be mowed close to the ground and maintained in a consistent direction because it affects the putting speed Many factors will affect the green index, and this part requires more research and testing The AI lawn mowing robot can identify and avoid obstacles through ultrasonic sensors while maintaining mowing quality The AI smart lawn mowing robot has a built-in camera that can be used to detect the health condition of the lawn Chao-Cheng Chen said that in the future, an expert system will also be introduced for early determination of whether there are diseases, pests in the lawn or whether there is sufficient moisture, and provide lawn health data analysis to customers, so that they can take preventive and response measures sooner to reduce disaster losses Chao-Cheng Chen, who is also a good golfer himself, said that golf has developed well in Taiwan However, due to weather factors, such as rainy and humid climate and typhoons, Taiwan's golf courses have harder soil and more potholes compared with top tier golf courses overseas If AI lawn mowing robots are to be widely introduced into golf courses, there are still many difficulties that must be overcome However, Taiwan's difficult terrain creates a good testing ground Once Taiwan can overcome the many problems and successfully introduce the robot, it will be able to expand to overseas markets and seize new market opportunities in a blue ocean Chao-Cheng Chen, President of USRROBOT nbsp