IDENTIFICATION OF THE NUMBER OF LOCULES IN TOMATO FRUITS (SOLANUM LYCOPERSICUM L.) USING COMPUTER VISION

FLAVIO E. SPETALE; VAZQUEZ DANA; JAVIER MURILLO; CACCHIARELLI, PAOLO

Biocell

Tech Science Press

Lugar: Henderson; Año: 2020 vol. 43

The locules are the internal cavities of the fruits that contain the seeds. In tomato (Solanum lycopersicum L.), the number of locules (NL) varies from two to fifteen or more. This character affects the final shape and size of the fruits, which are attributes of great biological, historical, and economic importance for the crop. Consequently, the determination of this trait presents agronomic importance and is relevant for breeding programs. Traditionally, the determination of the NL has been determined by visual counting from cross-sections of fruit. However, this method is time-consuming. On the other hand, sometimes the locules may be rudimentary or not fully developed, and it depends on the expert´s criteria whether or not they are considered in the count. This subjectivity generates noise in the data and makes it difficult to understand the mechanisms underlying these characteristics. In this work, an automatic method based on computer vision is presented for the identification and counting of NL in tomato fruits. Two data sets were used: the first includes 735 cross-sectional images of tomato fruits taken in the Villarino Experimental Field of the FCA-UNR and another available in the Sol Genomics Network repository consisting of 143 cross-sectional images. Both sets of images were taken following the same considerations. Edge detection was made using the automatic Otsu method and after the fruit was sectorized. An equalization histogram technique was then applied to homogenize the image and detect the internal edges. In the next step, the segmentation of the image based on an adaptive threshold was performed. Afterward, internal rings were generated in the image and within each ring, the number of locules was determined. Finally, a consensus value was obtained between the rings, making a mode. Promising experimental results were achieved in terms of accuracy in real tomato data sets, suggesting the possible usefulness of the proposed method in the development of cost-effective tools for automatic measurement of internal tomato attributes. A web interface for the automatic calculation of the NL from images of cross-sections of fruits was implemented.