スカイロジック List of Products and Services

If you have good and defective product images taken with a digital camera or similar, you can conduct detection tests using your own images. Please train DeepSky online and try out the detection performance of the AI. The detection results will be notified via email. 《1》Shooting: Take pictures of the acceptable and unacceptable products that you want to detect with a digital camera or similar, and save them in a single folder. 《2》Annotation: Click on the "Annotation" menu. Load the images you want to annotate from the folder in step 1, and place annotation boxes around the objects you want to detect in each image. 《3》Uploading Annotation Data: Click the "Learning" button in the "Annotation" section to upload the annotation data created in step 2 to the DeepSky program. After the learning is complete, a notification email will be sent to your registered email address. 《4》Detection and Recognition: Click on the "Detection and Recognition Test" menu and upload the images you want to detect and recognize to the DeepSky program. The result images will be sent to your registered email address as an email with image attachments.

We have received several inquiries from various food production and processing sites. We are now able to handle cases that were difficult for EasyInspector (formerly EasyInspector), such as detecting hair and foreign objects mixed in bento boxes, which further demonstrates the wide range of capabilities of our AI (deep learning) functions. DeepSky allows us to teach it with a broader scope regarding what we want to detect, making it possible to identify foreign objects that do not conform to a specific shape. If there are items that are currently being checked by human eyes and you wonder, "Can this be inspected?" please feel free to contact us.

Recently, we have been receiving inquiries asking if the inspection of agricultural products we are currently conducting can be made a bit easier. Using vegetables provided by local farmers, we tested whether we could: ■ Classify potatoes by grade ■ Detect damage on tomatoes using SkyLogic. ▼ Potato Classification We photographed potatoes classified from A to C from different angles (a total of 66 images were taken), and after creating the training data and conducting inspections, we were able to classify them into grades A, B, and C. (All 30 images were classified correctly, achieving a 100 percent accuracy rate.) ▼ Detecting Damage on Tomatoes Since it was difficult to find damaged tomatoes, we tested whether we could detect damage on one type. Similar to the potatoes, we photographed them from different angles (12 images), created training data, and conducted inspections, resulting in the detection of damage as shown in the images. (All 30 images were inspected, and all damage was detected, achieving a 100 percent accuracy rate.)

We have received a lot of inquiries from customers and have made improvements, resulting in an even more upgraded version compared to the initial release! We are now able to achieve higher precision counts for various objects. Before using cazoeTell for counting, it is necessary to create a learning model tailored to the objects you wish to count. 【Creating a Learning Model】 1) Capture teacher images 2) Perform annotation to create training data 3) Train the AI (create training data) ■Points to note when capturing teacher images ■About annotation ■Tips for annotation "If you're curious but unsure if you can do it," or "I'm having trouble with counting!" Customers interested in cazoeTell should first contact Sky Logic! We will create a learning model tailored to the objects you want to count. *For more details, please refer to the related links (blog).

The name "cazoeTell2" has been changed to "wakeTell." The reason for the renaming is that the name "2" often leads customers to mistakenly think it is a superior version of cazoeTell. cazoeTell is specialized in counting a single item and has high counting performance. wakeTell can classify and count multiple items, but its counting performance is lower than that of cazoeTell. ■ Comparison Table (Please refer to the related link (blog)) ■ Comparison of Counting Nuts and Washers Both cazoeTell and wakeTell can count a total of "30" items without any issues. However, cazoeTell only provides the result that there are "30" items without specifying how many of each item there are. On the other hand, wakeTell recognizes and displays the result as "15" nuts and "15" washers. Therefore, wakeTell can also be used for purposes such as forgetting to return tools.

EasyMonitoring, released in 2018, has been renewed alongside the image processing software "EasyInspector," and was re-released last autumn as EasyMonitoring2. Over the past year, we have introduced features that our customers have appreciated and new capabilities that have been added. With the addition of AI features... It has become possible to detect objects even in environments where monitoring was previously difficult due to significant changes. - Detection with stable accuracy even outdoors - Changes in brightness (trained in advance with images from various times of day and patterns) - The positions where objects appear are varied and different each time Each object can be identified and detected, allowing us to determine "where and what" has been detected. - Detection is acceptable or not based on specific areas Objects are detected with a sense close to that of the human eye. - Relative learning of objects with individual differences (animals, insects, agricultural products, foreign substances, etc.) - Detection of areas that cannot be judged solely by color (RGB) variations or roughness.

Until now, Skylogic has primarily handled image inspection projects focused on industrial products. This is because the nature of industrial products, which involves "mass-producing the same item," matched well with conventional (rule-based) image processing methods. Conversely, it has been difficult to handle items with unstable colors and shapes, even among industrial products, using traditional methods. However, with the emergence of image processing methods utilizing AI (deep learning), items with variability can now be included in the scope of image processing. This is because AI excels at "detecting only what we want to find while ignoring trivial changes." Now, connecting this to the title, when we talk about "items with unstable colors and shapes," we are referring to food products. Issues such as material defects, foreign object contamination, quantity defects, and missing components... even when the subject is food, products produced in factories are likely to face similar challenges as industrial products. However, the difficulty of processing with traditional methods has led to these issues being abandoned. (And I have also given up on them.) Why not try AI?

▼Obtaining Anisakis → Until Filming This time, we conducted a verification at our company to find "Anisakis," which we received several inquiries about from customers, using AI. We started by trying to obtain live Anisakis, but it turned out to be more difficult than we expected... We visited several supermarkets and fish shops in the city and made phone calls to request their help. Despite it being an extremely busy time of year, I made a strange request to unknown housewives, saying, "Can you give me live Anisakis for an experiment to find Anisakis with AI (to summarize)?" I am very grateful to the supermarkets that cooperated with us, and to the staff in the fresh fish section who helped us. Thank you very much. We assembled the equipment assuming an operation where we conduct AI Anisakis inspections on a slowly moving conveyor, and if detected, a buzzer would sound to stop the conveyor. Anisakis glows in response to specific wavelengths of ultraviolet light, so we use UV LED light sources and filters that pass specific wavelengths. In image processing, it is also important to capture images that make it easier to find Anisakis through optical manipulation.

EasyInspector2 has three text recognition features. 1) OCR (Optical Character Recognition) 2) Machine Learning OCR 3) AI OCR ■Simple setup ■Can read characters that were previously difficult to recognize ■Finally *For more details, please see the related link (blog).

Do you have any of these concerns? ■ It takes a lot of time for manual work ■ Grade classification is difficult, leading to oversights in visual inspections ■ Relying on hired labor increases labor costs ■ While skilled workers are aging, there are no successors Automate with image inspection software! 1) The defect detection tasks that were done manually can be automated ⇒ Detect scratches, discoloration, dirt, etc., and reduce work time! 2) Difficult grade classification can be easily done by anyone ⇒ Various rules can be used for differentiation! Free yourself from manual work! 3) Operate from a single computer! Start at a low cost ⇒ Significantly reduce implementation costs! Keep the number of workers to a minimum and reduce labor costs! Prevent oversights in visual inspections and enhance quality value! You can start small without the need for large machinery! 【Inspection Examples】 ■ Classification of tomatoes by size ■ Classification of flower types ■ Identification of corn seeds ■ Differentiation of tea leaf types ■ Inspection of orange blemishes ■ Detection of defects in coffee beans *For more details, please see the related links.

With the development and spread of AI, it has become possible to automatically detect foreign objects (such as insects, frogs, anisakis, hair, etc.) in food that were previously considered difficult to identify. AI can discover foreign objects that are often overlooked during quick tasks. ■ Hair contamination in food ■ Confirmation of ingredients in cup noodles ■ Identification of bread types on trays ■ Differentiation of meat cuts ■ Counting packaged sweets ■ Inspection of egg cracks ▼ Preventing oversights with AI to ensure food safety Issues such as damage or rot in agricultural products, ingredient shortages in food, and foreign object contamination cannot be completely prevented even with visual checks, leading to oversights. AI continuously monitors agricultural products and food without fatigue, efficiently reducing the chances of missing anything. ▼ Standardization and labor-saving in sorting Agricultural products have various criteria for checks, such as damage and insect bites, and also need to be sorted based on color and shape. When done by human hands, there is inevitably some oversight and subjective judgment, making it difficult to maintain consistent quality. The AI's vision checks and sorts based on constant standards, contributing to the standardization and labor-saving of sorting processes.

Insects, plastic, frogs, anisakis, hair, and other foreign objects are monitored at a speed of about 10 times per second. AI eyes efficiently detect foreign objects on behalf of human eyes and notify with lights or buzzers. When connected to a PLC, it is also possible to stop or discharge the conveyor upon detection of foreign objects. ▼ Easy installation on conveyor belts Even in places where it's difficult to place a PC, the aiforce-mini can be installed and used on the side of the conveyor or control panel. ▼ Detectable foreign objects Insects, plastic pieces, frogs, anisakis, hair, and other items that can be visually confirmed. ▼ Beyond foreign objects Through learning, it can also check for scratches, damage, burns, and chips on the conveyor. 【Product Composition】 - aiforce-mini main unit (AI software pre-installed) - Industrial camera - Lens (specifications are selected based on the inspection target and shooting distance) - USB cable for the camera (3m) * A mouse, keyboard, and monitor needed for setup are optional.

Until now, the focus has primarily been on the utilization of equipment within factories and manufacturing sites, but with AI capabilities, it has become possible to detect humans and animals, allowing for applications in various industries such as agriculture. This contributes to reducing time and effort through verification. ■ Management of control panels in greenhouses Automates visual patrol checks using cameras and image processing. ■ Measures against wildlife damage Ignores humans and vehicles, detecting only birds and animals. ▼ Image collection possible in areas without power In forested areas and farmland without power, it is possible to supply power using solar batteries and collect images via mobile phone lines. Real-time notifications can be sent regarding damage from wild boars and deer, as well as illegal dumping. 【Functionality and monitoring examples】 Analog meter reading Reading flow meters, float meters, etc. Digital display (7-segment) numerical reading Lamp status (such as in control panels) Dot matrix numerical reading Reading values and displays from centralized management PC monitors Reading vertical rotation power meters and difficult-to-read characters Monitoring absence of people and entry into restricted areas Detection of insect contamination and animal intrusion Condition of liquids

Checking equipment located at a distance involves the hassle of moving and the risk of accidents during transit. Additionally, it is often difficult to check frequently, and it requires considerable effort for recording and management. ■ Control panel lamps, instruments, and switches ■ Monitoring pressure, temperature, voltage, and valves in the plant ▼ Automatic monitoring, recording, and early abnormality detection using network cameras An image processing system using network cameras can automatically read and record various equipment and meters located at a distance. By allowing for frequent checks, it enables the early detection and response to abnormalities. Furthermore, not only can it read meters, but it can also automatically detect abnormalities such as valve orientation, leaks, and turbidity of treated water through AI image processing. It can also be applied to sensory judgments that require human confirmation, comprehensively realizing the concept of "no need to go check" for patrol inspections.

The manufacturer we consulted this time is a company that has been inquiring with us for some time. They sent us sample images. Our company also conducts usage explanations and demonstrations via web conference. First, it would be smoother to enter the verification process after asking about the client's issues and operational status during the web conference. ■ Inspection Settings and Results We conducted learning and processing on the two types of images provided, one with spectral lighting and the other with general lighting. For both images, we specified abnormal areas and trained the system to identify them. In the spectral lighting image, we were able to detect some potentially defective areas, and I feel there is a possibility to improve detection accuracy by increasing the training data. In the general lighting image, it honestly seems difficult to make judgments. It appears challenging to detect anything other than large, grainy defects like those in the image on the right. The left image shows the detection of defects in the spectral lighting image. The numbers in the detection image represent the AI's confidence level in percentage, which we refer to as the number of recognition points.

In the mechanism for turning the 100V power ON/OFF created in 'Discovering Attached Insects and Stopping the Conveyor' (https://skylogiq.co.jp/DIY_HowTo/291), a large mechanical relay was used. By turning the contacts with a potential difference of 100V ON/OFF using the relay's electromagnetic coil, unexpectedly large electromagnetic noise was generated. Electromagnetic noise can cause communication failures and other issues in interfaces such as USB. This time, I used a solid-state relay (SSR) to create a power ON/OFF mechanism that minimizes electromagnetic noise. At the beginning of the video, a ring fluorescent light is connected to the 100V power, and the process is as follows: DeepSky makes an OK/NG judgment → OK/NG output to Intelligent I/O → the buzzer and solid-state relay connected to Intelligent I/O turn ON/OFF → the ring fluorescent light turns ON/OFF. If a conveyor belt is used instead of the ring fluorescent light, the conveyor will move when OK and stop (with the buzzer sounding) when NG, enabling automatic operation.

Even manufacturers of high-precision resin-molded parts, such as precision components, are considering our inspection software. In preliminary simple verifications before understanding the operational situation and requirements, we may report on both conventional rule-based inspection software and AI (deep learning) software to determine which better meets their needs. 【Inspection Settings and Results】 In the enlarged image of the detection by EasyInspector on the left, the field of view was set to accommodate the workpiece, and when inspecting defective samples, it was possible to detect black spots and dirt. However, if there are irregularities in shape or shadows of contours within the inspection area, there is a possibility of misdetection. Conventional software is more susceptible to positional deviations. The image on the right shows the case using DeepSky. Through deep learning, it learns defects and detects only the defective areas from the image. Due to its nature, if properly trained, it can detect the intended defective areas even in the presence of variations caused by pad adhesion. 【Software Used】 Software used: EasyInspector, DeepSky *The current EasyInspector2 also includes AI functionality.

We conducted a simple verification of the detection of scratches on metal parts. In our communication with the customer, we sometimes report both the traditional rule-based inspection software and the AI (deep learning)-based inspection software for their consideration of which is better. 【Inspection Settings and Results】 As a result of the verification using the samples provided, scratches on metal parts were detectable by using ring lighting. By utilizing the "Scratch Inspection" feature of EasyInspector, we were able to detect scratches in 0.34 seconds from one location (the entire screen). Adjustments to the settings are necessary for different diameters and lengths. DeepSky also achieved static detection in 0.3 seconds. The left image is an enlarged view of the detection area from EasyInspector. The right image is an enlarged view of the detection frame from DeepSky.

This is an inquiry about the setup method for measuring the distance between the centers of two circles. The inspection software "EasyInspector" that you inquired about is a user-friendly, general-purpose image inspection/image recognition software that can utilize low-cost USB cameras. You only need to install the software and connect a commercially available webcam. You can easily build an inspection system for under 100,000 yen. With over 1,000 successful implementations, it is used in various fields such as electrical, electronics, machinery, and assembly. 【Inspection Settings and Results】 By using the "Dimension Angle Inspection" function of EasyInspector, we were able to measure the distance between the center of one hole and another. The inspection was possible by setting two inspection frames, left → Inspection Frame 1 and right → Inspection Frame 2, over the holes and measuring the difference. This inspection is also a support for customers who are already operating our inspection software regarding the setup method. Our company accepts verification and support from technical staff on a daily basis. If you have any issues, questions, or uncertainties during operation, please feel free to contact us at any time.

This is a request for a simple verification from China. Our company has contracts with agents in China and Malaysia, and we are expanding sales locally. We may also introduce agents for inquiries related to the operations of Japanese companies' overseas manufacturing departments. This time, we will conduct a simple verification of the wear and tilt of the labels affixed to the products. 【Inspection Settings and Results】 Inspection was possible with EasyInspector. The left image shows the wear on the label and the settings for comparison inspection with the master image. The right image shows the angle of the label and the settings for the dimensional angle inspection function. By using this function, we were able to determine the wear and angle of the labels at three locations in 0.07 seconds. 【Software Used】 Software used: EasyInspector Current 'EasyInspector2' color package [comparison with master image] MS (MeaSure) package [position and width measurement] [angle measurement] can be used for inspection.

This is an inquiry from an industrial equipment manufacturer with whom we have previously done business. EasyInspector can be used for various parts and defects, but the accuracy of detection can vary depending on the settings. Although it is a one-time purchase inspection software, we will continue to provide support after implementation. 【Inspection Settings and Results】 Regarding the method of inspection using "color presence inspection," in this case of the reverse image, it is likely that detection was possible because it happened to be slightly elevated. Therefore, we would like to put the inspection method for "color presence inspection" on hold and ask you to consider the inspection method using "dimensional angle inspection." Even in the case of "dimensional angle inspection," if the position of the inspection item is elevated (upward on the screen), it may not detect the correct angle. An effective method is to add another inspection frame, set up color inspection and other settings there, and check whether the inspection item is not elevated too much.

This is a support case for a customer who has actually implemented DeepSky. We provided guidance to increase the detection rate. The "annotation" used to teach the AI where to detect is crucial for inspection accuracy. The image below shows annotations that had a low detection rate. We advised that annotations that are too large or too small decrease the detection rate. Our company offers verification and support from technical staff on a daily basis. If you have any issues, questions, or uncertainties during operation, please feel free to contact us at any time. 【Inspection Settings and Results】 We set the judgment to "fail" after detecting one or more defective areas. (The inspected images were unknown images different from the trained teacher images.) We inspected 32 unknown images (16 good products / 16 defective products) and achieved 30 correct judgments and 2 incorrect judgments (we failed to detect defects, and all good products were correctly judged). *We also inspected the teacher images, but there were 2 false detections. It is likely that increasing the number of training images and the training time will improve detection. 【Software Used】 Software used: DeepSky

We received an inquiry regarding the detection method of DeepSky from a pharmaceutical company we have been in contact with for some time. 【Test Settings and Results】 Image B is one of the four divided parts. The resolution is simply one-fourth. Assuming the area of defects in the entire field of view is 1, the area of defects in the entire field of view of Image B would be 4. Since the "target object" is significantly larger in the overall image, it becomes easier to detect in Image B, even though its resolution is lower. This is why the importance of resolution itself is lower in deep learning. 【Software Used】 Software used: DeepSky

The grain dryer manufacturer experienced the web trial of our inspection software DeepSky, which uses AI (deep learning) from our website, but did not obtain satisfactory inspection results. We provided guidance on the method for generating parameters. While personnel familiar with deep learning inspections can conduct inspections smoothly, it is natural that first-time users may not achieve the desired results. If you have any questions about unsatisfactory results or the settings, please feel free to contact us. 【Inspection Settings and Results】 We enclosed the target areas to be detected in frames and labeled them by type. This time, we used seven pieces of training data consisting of five types: the numbers 1 to 4 and one without a number. All images correctly identified the targets. 【Software Used】 Software used: DeepSky Number of inspection points: 2 locations, recognizing the numbers in two areas on the screen.

We will introduce common challenges heard from customers and ways to solve them. Challenge 1) It is difficult to analyze why a judgment was incorrect. Challenge 2) I want to quantitatively determine whether the completed model is good or bad. Challenge 3) When learning, I want to visually know which items are likely to be confused. Challenge 4) I don't know when to stop learning. When does overfitting begin? Challenge 5) Defective products are not being collected for new products. *For more details, please see the related link (blog).

The AI general-purpose appearance inspection software "EasyInspector2" has newly added the "AI Segmentation" feature. The "AI Segmentation" feature performs segmentation (region division and labeling of each pixel in the image) using deep learning within the image. It calculates the ratio of pixels detected by segmentation within a specified polygonal area and makes pass/fail judgments. This feature can mainly be used for the following applications: - Tank water level - Area of metal processing surfaces and coating area - Measurement of crop growth (size) - Detection of lesions in photos and measurement of their area - Checking for coating gaps in sealants and adhesives - Monitoring the amount of smoke emitted By using the AI Segmentation feature for judgment, it achieves: - Improved work efficiency: Significant time savings compared to visual inspection - Enhanced accuracy in quality control: Reducing human errors and stabilizing quality - Cost reduction: Lower labor costs and reduced costs due to suppression of defective products.

It can be combined with rule-based functions. When trying to measure the area or dimensions of a target object within a complex background, the background interfered, making it difficult to detect edges as desired. However, by extracting the target object and binarizing it using AI segmentation, inspections have become much easier. *For more details, please see the related link (blog).

Recently, we released a new product, the "AI Sheet Counter ai-Numbers." ai-Numbers is a new counting machine that uses AI to handle sheet materials where counting has been difficult due to variations in cross-sections caused by the state of cutting blades and significant differences in thickness and appearance based on the material. Today, we will introduce the differences between the counting machine "ai-Numbers" and "EasyNumbers." In manufacturing, "counting errors" directly lead to reduced yield and complaints. Image processing-based counting machines can resolve this issue at once, but the optimal solution varies depending on whether the edges of the counting targets are complex and difficult to understand or if the focus is on counting visually clear edges as quickly as possible. We will compare our counting machines, ai-Numbers and EasyNumbers, in terms of technology, application areas, and performance, and summarize the key points for selection. *For more details, please refer to the related link (blog).

In conventional image processing, it is possible to inspect the degree of terminal insertion, but there are challenges such as limited connector shapes and the need for positioning. Therefore, we decided to use AI for the inspection. By training the AI to recognize uninserted, partially inserted, and fully inserted states, it becomes possible to identify them. While the AI object recognition function is sufficient for "detection," the real issue is how to make the judgment. AI object recognition establishes inspection by having humans create settings that search for and recognize defects, such as scratches or dents, from the overall workpiece—if even one defect is found, it is deemed a failure (NG). In other words, object recognition alone can identify an object, but it cannot make a judgment that it is acceptable just because it appears in a certain alignment. This is where the OCR function comes into play. *For more details, please refer to the related link (blog).

Recently, there has been an abundance of inexpensive network cameras available, which has allowed our EasyMonitoring2 system (a system that collects images from network cameras for tasks such as meter reading) to be implemented at a lower cost. However, this has also brought about certain issues. Specifically: - Most cameras have become of a type that does not allow lens replacement. - The predominant lenses are those designed to cover a wide area, which often results in significant image distortion (similar to fisheye lenses). Due to these trends, it has become challenging to capture images of meters from a distance using inexpensive cameras. Since the EasyMonitoring2 system connects to a large number of cameras, there is a demand for compact and affordable options. Therefore, we decided to test a method that allows for capturing images of meters even with low-cost cameras. *For more details, please refer to the related link (blog).

This time, we used a PTZ camera with the aim of capturing multiple locations with a single camera. A PTZ camera is one that has the following mechanisms: Pan: to move the camera horizontally Tilt: to move the camera vertically Zoom: to zoom in (optically changing the angle of view). (Images are a general representation of PTZ cameras.) Let's quickly check the field of view. *For more details, please refer to the related link (blog).