Method for Classifying Tomatoes Using Computer Vision and MLP Networks
Quality control in a production line, when performed in a nonautomatic way, is subjected to mistakes made by the inspection agents, who over time and due to fatigue let products that are improper for human consumption pass through the mats.
- PDF / 281,964 Bytes
- 6 Pages / 439.37 x 666.142 pts Page_size
- 70 Downloads / 234 Views
Method for Classifying Tomatoes Using Computer Vision and MLP Networks
18.1
Introduction
Quality control in a production line, when performed in a nonautomatic way, is subjected to mistakes made by the inspection agents, who over time and due to fatigue let products that are improper for human consumption pass through the mats. The majority of automatic inspection systems is based on human vision and has a simulated artificial vision to perform product classification. Inspection systems with computer vision are used mainly on agricultural applications, which have a high correlation between food quality and its appearance. Automatic quality inspection of tomatoes has as its main motivation the dissatisfaction of consumers, as reported in several technical reports from the food industry. Therefore, the use of artificial neural networks for classifying tomatoes can make the inspection process more efficient, improving the quality of the products that reach consumers. The 102 tomato samples used for this application were acquired in supermarkets from the São Carlos region (Brazil). They belong to the “Saladete” group, whose standard pattern is rounded and a final reddish color. Figure 18.1 illustrates some tomatoes from this group. The tomato samples obtained had different colors and sizes and were grouped into four classes as shown in Table 18.1. Images from the tomatoes were done by a digital camera, using an acquisition module that also has a diffuse lightning camera, whose purpose is to reduce, maximally, unwanted reflections. The acquisition module can be seen in Fig. 18.2. In Fig. 18.3, an example of a digital image obtained by the module above is shown.
© Springer International Publishing Switzerland 2017 I.N. da Silva et al., Artificial Neural Networks, DOI 10.1007/978-3-319-43162-8_18
253
18 Method for Classifying Tomatoes Using Computer Vision …
254 Fig. 18.1 Tomatoes from the “saladete” group in different maturation phases
Table 18.1 Distribution of tomato classes
18.2
Classes
Equatorial diameter (cm)
Color
A B C D
Bigger than 7 Smaller than 7 Bigger than 7 Smaller than 7
Completely red Completely red Greenish spots Greenish spots
Characteristics of the Neural Network
From the digital images, it is possible to extract features that will be presented to the neural network for the learning process. From this process, six normalized variables were obtained, which represent the average pixels related to the colors Red (R), Green (G), and Blue (B), as well as the chromaticity coordinates of red (r), green (g), and blue (b). The chromaticity coordinates (r, g, b) are obtained from expression (18.1), while the average pixels (R, G, B)m are calculated as (18.2), that is: PN i¼1 ðR; G; BÞi PN PN R þ i¼1 i i¼1 Gi þ i¼1 Bi
ðr; g; bÞ ¼ PN
ð18:1Þ
18.2
Characteristics of the Neural Network
Fig. 18.2 Lightning and image acquisition module
Fig. 18.3 Example of an image acquired by the module
255
256
18 Method for Classifying Tomatoes Using Computer Vision …
PN ðR; G; BÞm ¼
ðR; G; BÞi ; N 255
i¼1
Data Loading...
