This page is for indexing purposes and will not be printed in the conference proceedings book
Author(s)
First Name / Middle Name / Surname / Role / Type (Corresp)Naoshi
Tomoo
Shigemune / Kondo
Shiigi
Taniwaki / Professor
Graduate Student
Associate Professor / Yes
No
No
Affiliation
Organization / URL / EmailDepartment of Mechanical Engineering
Ehime University, 3 Bunkyo-cho, Matsuyama, Ehime 790-8577, Japan, Phone: +81-89-927-9708, Fax: +81-89-927-9708 / (URL is optional) /
Author(s)
First Name / Middle Name / Surname / Role / Type (Corresp)Kazunori
Mitsutaka / Ninomiya
Kurita / Section head
Engineer / No
No
Affiliation
Organization / URL / EmailDepartment of Technology Development
SI Seiko Co., Ltd., 66 Takaoka, Matsuyama, Ehime 791-8036, Japan / (URL is optional) /
Author(s)
First Name / Middle Name / Surname / Role / Type (Corresp)Takao / Nishi / Engineer / No
Affiliation
Organization / URL / EmailD add ninth Co.,Ltd.1-2-24 Toyonari, Okayama, 700-0942, Japan / (URL is optional) /
Author(s)
First Name / Middle Name / Surname / Role / Type (Corresp)Kazuhiko / Namba / Associate Professor / No
Affiliation
Organization / URL / EmailFaculty of Agriculture, Okayama University, 1-1-1, Tsushima-Naka, Okayama, 700-8530, Japan / (URL is optional) /
Publication Information
Pub ID / Pub Name / Pub DateASABE will complete / Biological Sensorics / ASABE will complete
Construction of Virtual Low-Noise Space for Safety and Comfortable Work under Noisy Environment
N. Kondo[1], T. Shiigi[2], S. Taniwaki[3], K. Ninomiya[4], M. Kurita[5], T. Nishi [6], K. Namba[7]
Abstract
Many types of agricultural facilities for grains, fruits, nuts, and vegetables have been constructed at many places in the world. Although most of operations are mechanized in the facilities, high noises from the mechanical systems often force operators inefficient or unsafe work in the agricultural facilities. In this research, a virtual space where operators can smoothly communicate with low noises was created by noise canceling headphones, microphones, a TV camera, and a host computer. The headphones reduced noise by providing inverse phase of the noise. Operators’ speech signals including noise signals were collected by microphones at their mouths and were sent to the host computer from operators’ laptop PCs through wireless LAN. The operators’ positions on floor were measured by a TV camera installed on the ceiling of facility. The speech signals were sent back to individual operators’ headphones after the speech levels were amplified based on the operators’ positions and orientations, while the noises from the microphones were reduced by a method of spectral subtraction. From this system, an operator could hear the other operator’s speeches in the virtual low-noise space according to their distances and orientations.
KEYWORDS. noise, citrus fruit, machine vision, spectral subtraction, wireless LAN
Introduction
There are many types of agricultural facilities for grains, fruits, nuts, vegetables, in which mechanized and automated systems have been introduced. Although most of operations are mechanized, high noises from the mechanical systems often force operators inefficient or unsafe work in agricultural facilities. It is not rare that the noise level is over a criterion of the permissible range officially determined for operators under noisy conditions: 8 hours under 85dB (A), 4 hours under 88dB(A), 2 hours under 91 dB(A), an hour under 94 dB(A), and a half hour under 97dB (A). For example, although noise level from a tractor is about 90 dB(A) and there are places of noise level 80-105 dB(A) in a leek preprocessing and grading facility, operators often work for 7 hours or so.
It is said that the long-time working under such noisy conditions may cause the operators health problems such as hypacusis, circulatory deficit, visceral disease or other disease. In addition, it may make errors in operation or injury accident with increase of tiredness and loss of concentration. To prevent them, operators sometimes plug their ears, but the earplugs often block out their mutual communication and may expose them to danger.
Some researches to reduce the noises of agricultural machines by active noise control around operator have been reported (Peng et al., 1995; Yoshida et al., 1993). Recently, noise canceling headphones that can become a wearable power of hearing-assisting devices were commercialized and they are actually used in subway trains or in airplanes. In this research, a communication system with noise reduction was constructed and a virtual space was created with the noise canceling headphones, microphones, a TV camera, and a host computer through wireless LAN and a machine vision system which measured a distance between operators. Objectives of this research are to reduce noise at operators’ ears, to smooth mutual communication between operators, and to increase safety under the noisy conditions for improving operators’ working environment.
Noises in Fruit Grading Facilities
Grading Facilities and Noise Measurement Methods
Noises were analyzed to investigate noise level and frequency characteristics in several fruit grading facilities before designing a system to construct a low noise virtual space. An eggplant fruit grading facility, a leek preprocessing and grading facility, a tomato fruit grading system in a large scale tomato production greenhouse, and a citrus fruit grading facility in Japan were utilized as sample facilities. Noise levels were measured at about 100 points selected in the facilities by a noise level meter and contour maps of noise level were created. It was observed that the noises generated from many machines working in the facilities. The noises near the main noise sources were saved as wave files for a minute by an IC sound recorder. The noise frequency was analyzed by FFT (Fast Fourier Transformation).
Measurement Results and Discussions
Figures 1 to 4 show the contour maps of noise level, while figure 5 and 6 show frequency characteristics of the maximum noise sources and of the main conveyor at the facilities. Figure 1 shows measurement result of an eggplant fruit grading facility (Kondo et al., 2007). All sides of each eggplant fruit are inspected before and after turning over by special designed rotary tray with 6 color TV cameras and 4 monochrome TV cameras in a grading line in the facility. This grading system had 6 lines whose length was about 32 m, which consisted of 348 rotary trays. From figure 1, it was observed that conveyor noise was 87 dB(A) and the rotary tray noise was 84 dB(A). From the figures of frequency characteristics, noise level was high from lower frequency to higher frequency and the conveyor had a high noise level at 18 kHz compared with the other conveyors.
Figure 1. Eggplant Fruit Grading Facility (No.1).
Figure 2. Leek Preprocessing and Grading Facility (No.2).
Figure 3. Tomato Fruit Grading Facility (No.3).
Figure 4. Citrus Fruit Grading Facility (No.4).
Figure 2 shows a result of leek preprocessing and grading facility. This facility had two lines and each line included a leek skin peeling machine by use of high air pressure and a grading system through TV cameras. From the figure 2, it was observed that the peeling machine gave forth 104 dB(A) noise and that the conveyor noise was 84 dB(A). The peeling machine generated the highest noises, which had especially higher frequency components, among the machines measured in this time. The conveyor noise had features at 68, 95, 135, and 188 Hz.
Figure 3 shows a result of a tomato fruit grading system at a large scale tomato production greenhouse. Main noise sources were a sorting machine (81 dB(A)) before packing operation and a sorting line (76 dB(A)). The sorting machine generated a noise at 60 Hz, while the conveyor noises were at 74, 120, and 180 Hz.
Figure 4 shows a result of a citrus grading facility (Njoroge et al., 2002), where citrus fruits are inspected by 6 color TV cameras. About 70 operators work from 8:00 to 16:00 in this facility whose performance was 200 t/ day. Citrus fruits are conveyed by a roller pin conveyor (Kurita et al., 2006) and the fruits are turned over by roller pins rotation for all side inspection. The main noise sources were the conveyor (81 dB(A)) and the end of grading line (86 dB(A)), where roller pins were retracted and returned back to a place to start. The end of grading line noise had features at 300 and 700 Hz, while the conveyor noise had at 600 Hz. A circle in the figure is for a place for experiment described later.
From these figures, it was observed that the grading facilities had different noise levels and different frequency characteristics, because the machines handled different fruits and operations and they were set in different size houses and in different arrangements. It was considered that noise reduction method and system should be usable at the different type facilities. In this research, a method to construct a low-noise virtual space was investigated using headphones and microphones as wearable assisting devices.
Figure 5. Frequency Characteristics of the Figure 6. Frequency Characteristics of
Maximum Noise Sources. Conveyor Noises.
Low Noise Virtual Space
System Outline
Figure 7 shows a system to construct a low noise space under high noise conditions. An operator wears a headphone with active noise cancelling function so that most noises were reduced by providing inverse phase of the noise. Operator’s (speaker’s) speech was collected by a microphone (Mic. 1), but the speech signal contained noise signals collected by the same microphones at his/her mouth. Another microphone was set near the speaker’s ear to collect and subtract the noise from the speech signal with noise. This subtraction was done by the speaker’s portable PC and the speech signal was sent to another operator’s (listener’s) portable PC through wireless LAN.
The operators’ positions on floor were measured by a TV camera installed on the ceiling of facility. A PC for machine vision processed the acquired images and calculated the decrease rate based on the operators’ two dimensional positions and orientations on floor. The machine vision PC sent the decrease rate to the listener’s PC through wireless LAN. The listener’s PC created stereo signals so that the listener could hear the speaker’s speech. This system makes a low-noise virtual space where operators can hear the other operators’ speeches with reduced noises according to their distance and orientations.
Figure 7. Whole System for Construction of Low Noise Virtual Space.
Experimental Methods
Spectral subtraction. Speaker’s speech signals collected by Mic.1 contain noise signals collected by the same microphones. Mic.2 set near the speaker’s ear collects only noises. It is fundamentally possible to make only speech signals by subtracting the signals of Mic.2 from the signals of Mic.1. Spectral subtraction (Nomura et al., 2004) as described below was used in this experiment.
Signal from Mic.1 is expressed by equation (1).
where , , and are signal from Mic.1, speech signal, and noise signal from Mic.2 respectively. , , and are Fourier Transformation of , , and . Here, it is assumed that have no correlation with . Subtracted spectrum of speech signal of the speech signal is described as equation (3).
(3)
where a is subtraction coefficient and . a is used for controlling subtraction rate of Mic.2 from Mic.1. When a is small, noise reduction is not enough, while noise is mostly eliminated but speech may be distorted when a is large. There is a trade-off relation between noise elimination and speech distortion when a is set. Here, if is calculated, inverse Fourier transformed speech signal is expressed by equation (4).
(4)
As described before, noise is reduced when a is large, but some parts of speech are also removed and a distortion happens. To solve the problem, there are methods that a is changed in speech signal region and in noise signal region. Methods to predict speech and noise regions and to determine coefficient a are reported so far: a determination method according to segmental SNR, and speech and noise region prediction method by use of standard deviation . Since Mic.1 was set near operator’s mouth in this experiment, level of the speech signal is larger than that of noise and high RMS(root mean square) power is input to Mic.1. In addition, since Mic.2 was attached closed to operator’s ear position, the noise signal containing little speech signal is input. From these conditions, assuming that ratio of RMS power between speech and noise is constant, segmental SNR per frame (n = 512)is calculated as equation (5).
where and are RMS power of speech with noise and of noise respectively. Figure 8 shows actual speech signal and . From the figure, it was observed that signal was speech when value of was more than 0.2. 1 was assigned to a when was larger than 0.2 after calculation for a frame, while 0 was assigned to when was smaller than 0.2.
Figure 8. Discrimination between speech and white noise based on snr.
Sound level calculation methods according to distance and orientation. Decrease rate of speech signal according to distance between operators is expressed by equation (7) (Hiwatashi et al., 1987).
where r is distance between operators.
Decrease rate according to orientation is expressed by equations (8) and (9) (Masaoka et al., 2001).
where and are left and right decrease rates and x is amplitude ratio of speech signal. Here, q is speaker’s orientation and is expressed by counterclockwise angle (listener’s orientation is 0 (rad).) From these, final decrease rates for create left signal and right signal are described as equations (11) and (12).