I-tech Ab · OMXSTO:ITECH
I-tech Ab — Technical Presentation (September 2025)
17 slides from I-tech Ab's technical presentation, September 2025
Slide 1: Biofouling in-water assessment with ROV and machine learning based image recognition software
Biofouling in-water assessment with ROV and machine learning based image recognition software FILIPPO CASTELLI1, VERONICA PIAZZA1, ISABEL CARROZZO2, FRANCESCO ROSASCO2, ENNIO OTTAVIANI2 , FRANCESCA GARAVENTA1 1 CNR-IAS Institute for the Anthropic Impact and Sustainability in the marine environment (
Slide 2: Laboratory and semifield facility inside Genova Port
Introduction Laboatory and semifield falicity inside Genova Port -Material testing in marine environment -Atmospheric corrosion testing -Semi-field testing under controlled condition -Mesocosms -Laboratory Amphibalanus settlement bioassay -Laboratory Toxicity testing of paint leachates -Surface char
Slide 3: Biofouling related to planetary issues
Introduction Biofouling related to planetary issues Biofouling Planetary Boundaries • Invasive alien species • Greenhouse gas emission • Biodiversity loss • Climate change Johan Rockström (2009) International Antifouling Conference 2025 4
Slide 4: Introduction
Introduction Recently the IMO has published of the resolution MEPC.378(80) adopted on 7th July 2023 titled: “2023 GUIDELINES FOR THE CONTROL AND MANAGEMENT OF SHIPS’ BIOFOULING TO MINIMIZE THE TRANSFER OF INVASIVE AQUATIC SPECIES”. This IMO document poses the bases to harmonization of biofouling vis
Slide 5: Aim of the work
Introduction AIM of the work: Automize as much as possible in-water biofouling assessment with the aid of new tools like robotics and deep learning International Antifouling Conference 2025 6
Slide 6: Dataset: test panels
Methods Dataset : test panels To train the neural network on recognizing the presence and the type of fouling, images of paint samples from the CNR-IAS archive were elaborated. These image were available in large number with different lighting condition, at different stage of fouling coverage and wi
Slide 7: Dataset elaboration and masks production
Methods Dataset Original image Panel Mask elaboration and masks production These image were elaborated manually with to create 3 masks. The first highlight the total area of the panel, the second and the third highlight pixels corresponding to hard fouling and clear surface. The soft fouling is obta
Slide 8: Dataset: ROV image acquisition and treatment
Methods Dataset: ROV image acquisition and treatment an ROV with a high resolution camera (4K, 12Mpixel) was Images are elaborated to obtain masks for the used to capture videos of different type of hulls in database. The image is cropped to the central area different conditions where the lighting i
Slide 9: Deep Learning Segmentation
Methods Deep Learning Segmentation Each image was preprocessed by extracting the region of interest and resizing it to a fixed resolution of 512×512 pixels. For segmentation, we employed U- Net, a deep learning architecture designed to capture visual patterns and reconstruct them into a precise pixe
Slide 10: Segmentation Software Pipeline
Methods Segmentation Software Pipeline From the input image, a region of interest is extracted. For hull images, the central 512×512 portion is selected to capture the area with the highest quality. In the case of panels, a dedicated panel detector is applied, followed by resizing to 512×512 pixels.
Slide 11: Training results
Results Training results Image present in the database compared with the target mask obtained by an operator Legend: • Green: indicates the clear surface, • Blue: represents soft fouling, • Red: marks areas of hard fouling International Antifouling Conference 2025 12
Slide 12: Test results
Results Test results Image not present in the database compared with the target mask obtained by an operator Legend: • Green: indicates the clear surface, • Blue: represents soft fouling, • Red: marks areas of hard fouling International Antifouling Conference 2025 13
Slide 13: Discussion
Discussion Discussion • The model is responding positively recognising properly each type of fouling • We have an underestimation of hard fouling, probably because is less represented in the dataset ( we are currently adding more hard fouled images) International Antifouling Conference 2025 14
Slide 14: Perspectives: Graphical User Interface
Perspectives Perspectives: Graphical User Interface The user interface under development will allow users to upload a video and extract cropped frames from it, or alternatively upload individual images. These images are then cropped and segmented by the neural network into three fouling classes. Eac
Slide 15: Perspectives: Stitching feature
Perspectives Pespectives: Stitching feature The interface will allow users to download data tables, an automatically generated report, and images of the hull reconstructed by stitching the frames from video. International Antifouling Conference 2025 16
Slide 16: Acknowledgment: Robotics and AI for socio-economic empowerment
NextGenerationEU Aknowledgment: Robotics and AI for socio-economic empowerment The Ligurian innovation ecosystem, based on the scientific and technological domains of AI and Robotics P8.2.1 Automated underwater inspection of fouling on ships hull International Antifouling Conference 2025 17
Slide 17: Thanks for your attention
Thanks for your attention filippo.castelli@ias.cnr.it International Antifouling Conference 2025 18