magistrsko delo
Dominik Klanjšček (Avtor), Janez Perš (Mentor), Marija Ivanovska (Komentor)

Povzetek

V sladoledni industriji je poleg okusa zelo pomemben vizualni izgled, saj le-ta močno vpliva na prodajo določenega izdelka. Z namenom nadzora vizualne ustreznosti izdelkov so se razvili določeni slikovni standardi, ki jih zahtevajo kupci. V našem primeru smo si za slikovni standard zadali, da anomalije ne smejo biti večje od 5 mm, da jih smatramo kot sprejemljive. Ker je ročna kontrola posameznega izdelka zamudna naloga, so se razvili optični sistemi, s katerimi se ti slikovni standardi hitro in enostavno implementirajo. Zaradi visoke cene takih sistemov pa smo se odločili, da sami načrtamo sistem avtonomnega zaznavanja anomalij na izdelkih iz sladoleda, kar predstavlja bistvo tega magistrskega dela. Za dani problem je bilo treba najprej zajeti podatkovno bazo označenih slik izdelkov, za kar smo zgradili napravo z ustreznim ohišjem, osvetlitvijo in objektivom. Za naš sistem smo načrtali ustrezno ohišje, ki zadošča strogim standardom živilske industrije. Osvetlitev s svetlim poljem, ki ne ustvarja senc, z možnostmi nadgradenj. Senzor kamere smo izbrali tako, da je bil kompatibilen z izbranimi vgradnimi računalniki in dovolj občutljiv za zajem karseda ostrih slik. Za lečo kamere smo izbrali lečo z ultra širokim vidnim kotom, da smo imeli več fleksibilnosti pri postavljanju kamere. Lahko bi vzeli manjši vidni kot, s čimer bi povečali ločljivost izdelkov, vendar v fazi načrtovanja sistema nismo dobro vedeli, kako se bo napravo namestilo na industrijsko linijo. Za nadzor naprave smo uporabili tri vgradne računalnike, ki nadzirajo svetili, kamero in skrbijo za shranjevanje podatkov. Nastavili smo tudi številne parametre kamere za zajem karseda ostrega videa, iz katerega smo nato izrezali slike izdelkov, s pomočjo katerih smo zgradili testno in učno podatkovno bazo. Nastali podatkovni bazi smo morali pomnožiti z binarno masko, s katero smo omejili območje zanimanja, da smo se izognili problemu, kjer nam izbrana metoda detekcije anomalij smatra ozadje izdelka kot anomalijo. Na tako obdelani podatkovni bazi smo testirali enorazredno metodo PaDiM. Iz rezultata testiranja modela lahko vidimo, da je metoda sposobna uspešno ločiti med izdelki brez anomalij in vsemi ostalimi (ROCAUC = 0.975). Želeli smo poiskati prag, ki bi nam ločil izdelke z večjimi (nad 5 mm) in manjšimi anomalijami (do 5 mm), vendar nismo uspeli določiti takšnega praga, da bi bilo to izvedljivo. Do tega praga smo poskusili priti preko upragovljanja map intenzivnosti anomalij, a se je izkazalo, da ni proporcionalen velikosti anomalije posameznega izdelka, zato je bil poskus neuspešen. Z oceno maksimalne napake, ki jo PaDiM pripiše posamezni sliki, pa si prav tako nismo mogli pomagati, ker odraža le, ali gre za sliko izdelka z napako ali brez nje in nič ne pove o velikosti same anomalije. Intenziteta anomalije, zaznana z modelom PaDiM, je namreč odvisna tako od kontrastnosti anomalije kot tudi od velikosti anomalije. Na podlagi te ugotovitve lahko zaključimo, da je model PaDiM neprimeren model za določevanje, ali je določen izdelek še sprejemljiv glede na določeni vizualni standard. Z njim smo sposobni dobro doreči le, ali je izdelek brez napak ali ne.

Ključne besede

računalniški vid;optično zaznavanje anomalij;globoke nevronske mreže;vrednotenje detekcije anomalij;PaDiM;magisteriji;

Podatki

Jezik: Slovenski jezik
Leto izida:
Tipologija: 2.09 - Magistrsko delo
Organizacija: UL FE - Fakulteta za elektrotehniko
Založnik: [D. Klanjšček]
UDK: 004.93(043.3)
COBISS: 200115459 Povezava se bo odprla v novem oknu
Št. ogledov: 65
Št. prenosov: 15
Ocena: 0 (0 glasov)
Metapodatki: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Ostali podatki

Sekundarni jezik: Angleški jezik
Sekundarni naslov: Automated detection of anomalies in ice cream products
Sekundarni povzetek: In the ice cream industry, in addition to taste, visual appearance is very important because it has a strong influence on the sale of a certain product. To control the visual adequacy of products, certain image standards have been developed that are required by customers. In our case, we set for the image standard that anomalies should not be larger than 5 mm to be considered acceptable. Since manual control of individual products is a time-consuming task, optical systems have been developed. Utilizing the latter, these image standards can be implemented quickly and easily. Because of the high prices of such systems, we decided to design a system for the autonomous detection of anomalies in ice cream products, which represents the essence of this master's thesis. For the issue in question, it was first necessary to gain a database of marked product images, for which we built a device with suitable housing, lighting, and lens. For our system, we designed appropriate housing that meets the strict standards of the food industry and shadowfree brightfield lighting with upgrade options. We chose the camera sensor so that it was compatible with selected embedded computers and was sensitive enough to capture a series of sharp images. For the camera lens, we chose an ultra-wide angle lens to enable more flexibility in camera placement. We could take a smaller viewing angle, which would increase the resolution of the products, but in the system planning phase, we did not know exactly how the device would be installed on the industrial line. To control the device, we used three builtin computers that control the lights and the camera and take care of data storage. We also set several camera parameters to record as sharp video as possible. From the latter, we cut out images of products through which we built a test and training database. We had to multiply the resulting database with a binary mask by which we limited the area of interest to avoid the problem where the chosen anomaly detection method considers the background of the product as an anomaly. We tested the one-class PaDiM method on the database processed in this way. Based on the model testing results, we can see that the method can distinguish successfully between products without anomalies and all others (ROCAUC = 0.975). We wanted to find a threshold that would separate products with larger (over 5 mm) and smaller anomalies (up to 5 mm). However, we were unable to determine such a threshold that would be feasible. We tried to reach this threshold through the application of anomaly intensity maps unsuccessfully. This threshold turned out to be not proportional to the size of the anomaly of an individual product. We also could not use the assessment of the maximum error that PaDiM attributes to each image because it only reflects whether it is an image of a product with or without an error and does not say anything about the size of the anomaly. Namely, the intensity of the anomaly detected by the PaDiM model depends on both the contrast of the anomaly and the size of the anomaly. Based on this finding, we can conclude that the PaDiM model is an inappropriate model for determining whether a certain product is still acceptable according to a certain visual standard. With it, we can only determine whether a product is defect-free or not.
Sekundarne ključne besede: computer vision;optical anomaly detection;deep neural networks;evaluation of anomaly detection;PaDiM;
Vrsta dela (COBISS): Magistrsko delo/naloga
Študijski program: 1000316
Konec prepovedi (OpenAIRE): 1970-01-01
Komentar na gradivo: Univ. v Ljubljani, Fak. za elektrotehniko
Strani: 1 spletni vir (1 datoteka PDF (X, 60 str.))
ID: 24473701