Positive results from the first phase of the project “Vision and Artificial Intelligence for hide grading”

The feasibility of using this technology for grading hides has been demonstrated.

This September the cluster organised a session at Adoberia Bella to present the results and conclusions of the project Vision and Artificial Intelligence for hide grading, a project led by Leather Cluster Barcelona in which three member companies took part: Intarex, Picvisa and Curtidos Badia. The project received funding from ACCIÓ (Ministry of Business and Labour), within the framework of the competitive strengthening grants for members of the Catalunya Clústers Programme 2021-2023.

The grading of hides is a process that today is carried out manually, based on the skill of workers with the aptitude to perform the operation for hours on end, using mental patterns learned through accumulating a great deal of experience, in order to distinguish between different types and combinations of defects that allow them to grade each hide into different categories. It is an operation of great importance for its economic impact on the business, given that the cost of hides as a raw material is usually around 50% of the total manufacturing cost, and producing a batch with inaccurate grading is detrimental to the quality of the manufactured product, customer satisfaction and the operating results of tanneries.

The purpose of this collaborative project, with innovation as its focus — applying technology to hide grading — is to open new doors to knowledge and business by providing tangible evidence about what is bought and what is produced.

The current phase of this project has been aimed at studying the technical feasibility of an automated hide grading system based on the application of vision and artificial intelligence.

This first technical feasibility phase has focused on the automatic identification, through images, of a set of defects in wet-blue hides, the state in which the hide is already half tanned and can be graded in order to assign it to different colours and finishes. Specifically, 27 different types of defect were characterised and 7 were selected for their importance in terms of impact and relevance to quality. Some 500 images were collected, reviewed and labelled for each identified defect category. A total of 3,600 images classified into 8 different categories (seven of the characterised defects and one of defect-free hide).

To carry out the project, a hardware and software prototype for capturing images of hides was designed, built and put into operation. Machine vision models and a minimum viable equipment set (MVP) for detecting the different defects were also built, evaluated and validated.

In the trials implementing Vision and Artificial Intelligence technology in hide grading carried out in this first phase of the project, accuracy rates of 84% were obtained, a figure higher than initially expected. It is estimated that with a larger image database the accuracy rate could exceed 90%.

These very positive results encourage the management of Leather Cluster Barcelona to proceed with a second phase of this project, which would be aimed at experimental development and the construction of equipment that can be integrated into automatic processing lines with the aim of analysing whole hides automatically and producing a detailed quality map, in order subsequently to apply the knowledge acquired to finished leather as well.

Leather Cluster Barcelona is a competitive cooperation environment for promoting cross-cutting, transformative strategic projects with the aim of driving forward the ecosystem of the Catalan leather sector. The cluster is an accredited body within the Catalonia Clusters programme and has the support of Acció.