Details

Autor(en) / Beteiligte

• Pariente, Enrique S.
• Cancilla, John C.
• Wierzchos, Kacper
• Torrecilla, José S.

Titel

On-site images taken and processed to classify olives according to quality – The foundation of a high-grade olive oil

Ist Teil von

• Postharvest biology and technology, 2018-06, Vol.140, p.60-66

Ort / Verlag

Amsterdam: Elsevier B.V

Erscheinungsjahr

2018

Quelle

Alma/SFX Local Collection

Beschreibungen/Notizen

• [Display omitted] •Novel tool to seek a specific quality in raw materials as well as in final products.•Combination of image processing and neural network to classify fruit ripeness.•Accurate tools for classifying olives of a specific quality grade.•Simple way to obtain the best olive oil quality by selecting the best olives. It is well known that the quality of the olives used to produce olive oil is directly related to the quality of this product. In this research, the classification of olives according to their quality grade has been achieved by combining image processing and mathematical modeling. The images of 190 different olives of four varietals (Ascolano, Manzanillo, Mission, and Sevillano), which were categorized into four quality-based groups (optimal, acceptable, borderline, and unacceptable), were employed to design a set of linear (partial least squares) and non-linear (artificial neural networks) mathematical models. Three different binary classifications were attempted: (a) optimal/others, (b) optimal and acceptable/others, and (c) unacceptable/others. The results obtained for (a) were a perfect classification rate using either one of the modeling approaches. On the other hand, (b) and (c) were clearly favored by the use of neural networks, which reveals a more complex task (images are more similar). In these two classifications, the linear models offered 75% and 70% correct classification rates, whereas the non-linear ones provided 93% and 90%, using comparable validation procedures. In addition, an external validation has been done using photographs taken directly from the olive grove. The misclassification percentage is less than 13%. Therefore, proper data extraction from olive images combined with neural networks can lead to accurate tools for the location of olives of a specific quality grade, which can be of great use for olive oil producers that seek a specific quality in their raw materials and, consequently, in their final product.