^{1}

^{1}

^{1}

^{1}

^{2}

^{2}

^{2}

^{3}

^{3}

This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

Electronic canopy characterization is an important issue in tree crop management. Ultrasonic and optical sensors are the most used for this purpose. The objective of this work was to assess the performance of an ultrasonic sensor under laboratory and field conditions in order to provide reliable estimations of distance measurements to apple tree canopies. To this purpose, a methodology has been designed to analyze sensor performance in relation to foliage ranging and to interferences with adjacent sensors when working simultaneously. Results show that the average error in distance measurement using the ultrasonic sensor in laboratory conditions is ±0.53 cm. However, the increase of variability in field conditions reduces the accuracy of this kind of sensors when estimating distances to canopies. The average error in such situations is ±5.11 cm. When analyzing interferences of adjacent sensors 30 cm apart, the average error is ±17.46 cm. When sensors are separated 60 cm, the average error is ±9.29 cm. The ultrasonic sensor tested has been proven to be suitable to estimate distances to the canopy in field conditions when sensors are 60 cm apart or more and could, therefore, be used in a system to estimate structural canopy parameters in precision horticulture.

Ultrasonic sensors have been used for many purposes in agriculture for more than 40 years. One of these applications has been the detection and ranging to extract geometric information from fruit tree canopies. The first developments in this area were related to the application of plant protection products such as pesticides and fungicides in fruit orchards. Once doses started to be adjusted according to the amount of vegetation to be treated [

Most of the referenced applications of ultrasonic sensors in canopy characterization are focused on correlating manual estimations of width or volume with the results obtained by using the sensors. These works did not provide information about the interaction between the sound wave and the canopy itself and how can it interfere in the estimations of ultrasonic sensors. In ranging applications, ultrasonic sensors are intended to estimate distances to objects with solid surfaces in order to get specular reflections of the acoustic waves. The surface of an apple tree canopy is made up of small leaf surfaces placed in different orientations. Consequently, the reflection of ultrasonic waves is more diffuse than specular and this can strongly affect estimated distances.

In this paper a commercial ultrasonic sensor model is analyzed to validate its suitability, performance and reliability in an apple orchard and for a better understanding of the sensing process. Trials have been performed in laboratory and field conditions in a stationary way in order to assess its ability to estimate both distances to the canopy and the effect of possible interferences coming from other adjacent sensors. This study is the first step in a larger work that would use this type of sensors to estimate canopy parameters such as cross section areas or canopy volume in fruit tree crops in the framework of precision horticulture.

One ultrasonic sensor and a data acquisition system were used in the laboratory and field trials. The sensors were mounted on a vertical aluminum mast on a mobile platform, which was also used to carry the acquisition system and the batteries to supply the required voltage to run the electronic devices.

The ultrasonic sensor selected to carry out the trials is a Sonar Bero PXS400 M30 K3 (Siemens AG, Munich, Germany,

The acquisition system (

An important parameter to be taken into account is the beam angle. In the specification sheet delivered together with the sensor, the information provided related to this parameter is as follows:

Ultrasonic sensors are also very sensitive to interfering sonic waves coming from nearby sensors. The manufacturer offers two methods to synchronize several sensors in order to avoid inaccurate readings due to interferences of sonic echoes sent from another sensor. However, both methods present big disadvantages. In one method the object cannot be assigned to a particular sensor. In the other, longer response times are required because each sensor is only active briefly and then has to wait until all the other sensors in the circuit have emitted. The latter, causes the array of sensors not to obtain simultaneous readings, what could cause inaccurate estimations of cross-sectional areas of tree rows. These solutions could be useful in a stationary system but are not satisfying at all when the sensors are boarded in a moving platform. In this last situation, a minimum distance between active sensors must be found in order to obtain the highest vertical resolution to better estimate canopy parameters with the least effect of interferences.

The laboratory distance measurement trial was carried out at the

The statistical analysis consisted of fitting a linear regression model (^{2}), the root mean square error (RMSE) and the significance of the model.
_{0} is the estimation of the intercept; β̂_{1} is the estimated parameter multiplying the regressor; v is the independent variable, that is, the sensor output.

Both distance measurement and interference trials were conducted in an

The aim of this trial was to compare the analog output of the sensor when measuring different distances to the canopy with the output when measuring the distance to an ideal target. To this purpose, a methodology and a test bench were designed and implemented (

The statistical analysis of the field data is analogous to the one carried out for the laboratory trial in terms of the type of fitted model (

The aim of the trial was to assess the effect of adjacent sensors working simultaneously with the central one. The manufacturer suggests that the user experimentally determine the distance between sensors to avoid interferences in situations other than specular sonic reflection. The interference trial has been carried out with the same test platform used in the distance measurement trial. In this case, the platform was fitted up with two more pairs of ultrasonic sensors placed at ±30 cm and ±60 cm around the central ultrasonic sensor in the vertical plane (

These distances have been chosen considering possible future applications of the sensor related to canopy characterization. In order to better estimate parameters such as cross sectional canopy areas or canopy volume, the closer the sensors could be the higher the vertical sampling resolution. In the previously referred analysis carried out in the citrus grove [

To this purpose, the methodology described in the flux diagram of

The total number of observations was 34,000. However, fifteen of them were removed as they were considered outliers. In

The regression line, as well as results of the distance measurement laboratory trial, is shown in

Total number of observations was 168. Ten observations were removed as they were considered abnormal when compared with their equivalent measurement to the 10 cm^{2} cardboard target. In

The average absolute error is 5.11 cm. A frequency analysis of the residuals in absolute value shows that the 20.9% of observations have an absolute error smaller or equal to 1.5 cm, when compared with the regression line established by the fitted linear model. A 52.5% of observations have an absolute error smaller or equal to 4 cm, a 74.7% smaller or equal to 6 cm and a 91.1% smaller or equal to 10 cm (

In order to compare the model for laboratory conditions with the one obtained in field conditions, a numerical simulation has been carried out. In

However, canopy surface characteristics provoke a higher variability in distance estimations. This is caused by the capacity of leaves to generate a sufficient echo for the sensor to acknowledge it. In

Most times a single leaf is not enough to generate a sufficient echo due to its reduced dimensions and/or to its orientation. To estimate a correct distance, the sensor needs to detect a group of leaves approximately placed in the same vertical plane (

If this group of leaves is not enough to produce a suitable echo, it will not be produced there, but rather some distance further away where a bigger mass of leaves can generate one (

Alternatively, it is possible that canopies present gaps in the outer layer of leaves. When a sensor is aligned with one of these gaps it could happen that the echo is produced by the surrounding leaves of the gap instead of by the leaf mass inside the gap (

A total of 113 observations were carried out according to the methodology described in the flux diagram in

In

According to the results and to the information provided by the manufacturer, the more apart the sensors are, the lower the effect of interferences is. However, a balance should be found between the sensor separation and the highest vertical resolution (maximum number of sensors to estimate the canopy vertical outline in a more accurate manner). Anyhow, it would be possible to overcome the effects of interferences by using an appropriate filter,

The tested ultrasonic sensor is able to accurately estimate distances under laboratory conditions with an average error of ±0.53 cm. When used under field conditions, the distance estimation equation should be adapted to better estimate distances to the canopy. However, differences with the laboratory estimation equation are relatively small, considering other possible sources of error.

The variability in distance estimations in field conditions in an apple orchard clearly increases in relation to what was obtained in laboratory with artificial targets. As a consequence of this, the average error is ±5.11 cm. The effect of interferences is higher when sensors are 30 cm apart with an average error of ±17.46 cm. When sensors are separated 60 cm, the average error is ±9.29 cm. Sensors should thus be separated more than 60 cm in order to avoid high interference effects.

Ultrasonic sensors like the one tested and reported in this paper have been proven to be suitable to estimate distances to the canopy in field conditions. Results could be extrapolated to other apple crop varieties and other species such as pear crops where canopy structures and leaf dimensions are similar.

However, it has to be taken into account that the increase of variability due to the characteristics of the canopy surface and the ultrasonic working principle reduces the accuracy of the estimations and that the effect of interferences can be important when adjacent sensors are too close.

The authors wish to thank Pere Masana, Xavier Torrent, Francesc Tolós, Josep Maria Vallès and Pere Fontbuté for their collaboration in the preparation and execution of the trials and Jaume Arnó for his suggestions on the statistical analysis. This work has been funded by the Spanish Ministry of Science and Innovation and by the European Union through the FEDER funds and is part of research projects Pulvexact (AGL2002-04260-C04-02), Optidosa (AGL2007-66093-C04-03) and Safespray (AGL2010-22304-C04-03).

Ultrasonic sensor used in this work (

Sound cone diagrams for different types and orientations of targets with a superimposed 5° beam angle cone projection in red.

Experiment layout for the laboratory distance measurement trial.

Test platform designed for both the field distance measurement and the interference trials (

Flux diagrams of the programs designed for the field distance measurement (

Scatter diagram of sensor output and distances (

Scatter diagram of sensor output and distances to the first leaf (

Histogram and cumulative frequency of residuals in absolute value for the field distance measurement trial.

Interaction possibilities between ultrasonic waves and canopy.

Scatter diagram of central sensor alone output and simultaneously working with adjacent sensors at ±30 cm (

Histogram and cumulative frequencies of errors caused by interferences from adjacent sensors at ±30 cm and at ±60 cm for the field interference trial.

Specifications of the ultrasonic sensor used in this work (model Sonar Bero PXS400 M30 K3, Siemens AG, Munich, Germany).

Sensing range | 40 cm to 300 cm |

Target dimensions for max. meas. dist. | 5 cm × 5 cm |

Response time | 50 ms to 200 ms |

Accuracy | ±1.5% |

Resolution | 1 mm |

Beam angle | Approx. 5° |

Sensor output | 0 VDC a 10 VDC |

Ultrasound frequency | 120 kHz |

Weight | approx. 150 g |

Ambient temperature (compensation) | −25 °C to +70 °C |

Operating voltage | 20 VDC to 30 VDC |

Vibrating stress | 11 to 55 Hz, 1 mm amplitude |

Shock stress | 30 g, 18 ms |

Degree of protection | IP 65 |

Characteristics of the

Phenological stage | BBCH 76 |

Row spacing | 5.00 m |

Tree spacing | 1.60 m |

Representative canopy width | 1.75 m |

Representative tree height | 3.75 m |

Summary of the fitted linear model to estimate distances to an artificial target from the sensor output in laboratory conditions.

Observations | 33,985 |

Coef. of determination (^{2}) |
0.999 |

Coef. of correlation |
−0.999 |

RMSE (cm) | 0.699 |

p value of the model | <0.0000 |

Summary of the fitted linear model to estimate distances to the first leaf from the sensor output in field conditions.

Observations | 158 |

Coef. of determination (^{2}) |
0.980 |

Coef. of correlation |
−0.990 |

RMSE (cm) | 8.107 |

p value of the model | <0.0001 |

Numerical simulation comparing distance estimations according to field and laboratory measurements.

0.00 | 294.28 | 297.07 | −2.79 |

1.00 | 268.46 | 270.71 | −2.26 |

2.00 | 242.63 | 244.36 | −1.72 |

3.00 | 216.81 | 218.00 | −1.19 |

4.00 | 190.98 | 191.64 | −0.66 |

5.00 | 165.16 | 165.29 | −0.13 |

6.00 | 139.34 | 138.93 | 0.40 |

7.00 | 113.51 | 112.58 | 0.94 |

8.00 | 87.69 | 86.22 | 1.47 |

9.00 | 61.86 | 59.86 | 2.00 |

10.00 | 36.04 | 33.51 | 2.53 |