• Portada - Success stories: First installations of IngiGrader 2.0 for external quality classification in citrus
Clementina calidad superextra

Success stories: First installations of IngiGrader 2.0 for external quality classification in citrus

We have successfully installed the latest version of our external quality detection system during the first part of the citrus season at several of our closest customers.

Despite the difficulties during 2020 due to COVID19 that caused a significant delay in the final development of the system, we were able to install some units in clients close to our offices.

Our system for detecting the external quality of the fruit is totally based on the use of self-learning methodologies based on neural networks. After several years of development, our technical department was able to optimize the architecture used, based on deep convolutional and recurrent networks, until achieving a highly accurate, robust system that was fully adaptable to the particular needs of each customer.

We decided to focus our efforts on adapting the system for the work of the clementine varieties that are worked in the period between September and December in the Spanish East area. We have a dataset of 500,000 examples on which we have trained the network that were manually classified by our technicians under the criteria of our most demanding clients. We decided to make a classification based on «quality grade» instead of orienting the system to the detection of defects or particular pests. In this way, the use by the end customer of the system is very simple, you only have to choose the degree of quality you want for the fruit based on the necessary classification. However, this gradation can at any time be modified on demand to include particular defects in some of the levels.

Good oranges
1) Fruits without skin or shape defects
Defect oranges
2) Fruits with skin defects, shape or incomplete coloring

Fruit selection carried out automatically by the system on a batch manually preselected by the plant operators and assigned as the first category.

We have applied the system to obtain different objectives: selection of the super-extra product from the total set of fruit for the preparation of highly demanding orders, division of the product previously selected as second quality into three additional levels, discard of fruits with extreme defects both skin-like or obtaining intermediate classes. Our system is permanently under development, while the system is working it saves data and fruit examples that will later be evaluated by a technician to reinforce or adapt the training to the particular customer.

Good oranges 2
Image 1
Defect oranges 2
Image 2

Quality extremes automatically graded by the system on a batch of fruit preselected by the plant operators and assigned as the second category. An attempt was made to «rescue» those fruits 1) that only had small skin defects or loss of the peduncle (stalkless) and on the other hand 2), those fruits with large defects or malformations due to hemipters (Planococcus citri) that would be of a category below the one originally assigned.

Our line of work is to be able to offer a system with the best value for money, always under the premise that it is economically profitable for the customer, with a very high rate of return. Although today it is impossible to totally eliminate handwork in the external quality selection tasks, our system helps to optimize the process and reduce the associated production costs, it also allows to establish a constant quality standard for the customer, regardless of factors. external or subjective.