Aerospace and Electronic Systems - August 2018 - 64

UAV System for Photovoltaic Plants Inspection
Table 1.

Precision, Recall and F-Measure Metrics for Panel and Defects Detection, Obtained for Different Thresholds
Panel Detection
Dataset

Threshold

Sequence 1
(86 images)

Sequence 2
(184 images)

Precision

Recall

F-Measure

0,15

0,10

1,00

0,25

0,13

0,35

64

tp
tp + fp

Time
(sec)

Precision

Recall

0,18

0,00

1,00

0,01

1,00

0,22

0,01

1,00

0,01

0,17

0,98

0,28

0,01

1,00

0,02

0,45

0,25

0,95

0,40

0,02

1,00

0,03

0,55

0,40

0,93

0,56

0,03

1,00

0,06

0,65

0,46

0,93

0,61

0,06

1,00

0,11

0,75

0,56

0,91

0,69

0,11

1,00

0,19

0,85

0,63

0,89

0,74

0,60

0,75

0,67

0,95

0,80

0,86

0,83

0,75

0,75

0,75

0,15

0,10

1,00

0,18

0,02

1,00

0,05

0,25

0,13

1,00

0,22

0,04

1,00

0,09

0,35

0,17

0,96

0,28

0,12

0,93

0,21

0,45

0,25

0,93

0,39

0,24

0,93

0,38

0,55

0,40

0,90

0,55

0,33

0,90

0,49

0,65

0,46

0,86

0,60

0,53

0,87

0,66

0,75

0,56

0,84

0,67

0,81

0,87

0,84

0,85

0,63

0,82

0,71

0,83

0,83

0,83

0,95

0,81

0,79

0,80

0,85

0,77

0,81

analyzed. Black boxes with a G-number ID are used to identify
panels on the PV plant. The panels that are contained in the images are correctly and uniquely identified as shown in the first
row of images in Figure 7, while some panels are missing in the
other case, shown in the second row of the same Figure. Blue
boxes with a B-number ID are used to identify defects on the PV
plant panels. Two heating points (IDs: B5 and B9) are correctly
detected, even if of different dimensions, and the false positive
(ID: B13) present in both (a) and (b), is then corrected in the
subsequent shot (c).
Three metrics have been defined and used to evaluate the classification results: Precision, Recall, and F-Measure.
The precision has been computed as the proportion of the examples that truly belong to a specific class among all those which
were assigned to that class. It is the ratio of the number of correct
items that are detected by the total number of irrelevant and relevant items detected:
Precision =

Defect Detection

(2)

45 s

108 s

F-Measure Time (sec)

28 s

55 s

where tp indicates the number of true positives and fp indicates the
number of false positives.
The recall has been computed as the proportion of items that
were assigned to a specific class, among all the items that truly
belong to that class, i.e., how much the class has been covered.
It is the ratio of the number of relevant items detected to the total
number of relevant items in the search space:

Recall =

tp
tp + fn

(3)

where fn indicates the number of false negatives.
The F-Measure is a measure of test accuracy, interpreted as a
weighted average of the Precision and the Recall parameters above
specified:
F-Measure 2 

IEEE A&E SYSTEMS MAGAZINE

Precision  Recall
Precision  Recall

(4)

AUGUST 2018
Aerospace and Electronic Systems - August 2018

Table of Contents for the Digital Edition of Aerospace and Electronic Systems - August 2018
