Reprinted from: The North American Journal of Fisheries Management 14.650-655, 1994 American Fisheries Society
EFFICIENCY AND ACCURACY OF AN AUTOMATED DATA CAPTURE AND ERROR-CHECKING SYSTEM FOR LABORATORY FISH PROCESSING
C. BOURQUE AND D.K.CAIRNS
Science Branch, Department of Fisheries and Oceans
Box 5030, Moncton, New Brunswick E1A 6E8, Canada
Abstract - An automated data capture system comprising an electronic measuring board and digital balance linked to a microcomputer was developed for laboratory fish processing. The system conducts data plausibility tests on lengths, weights, gonad weights, and gonad maturity states in real time, which allows errors to be corrected before the fish is discarded. In trials with samples of Atlantic herring Clupea harengus, the system detected data entry errors for 2.4% of fish; without the verification system, these errors would have entered data files. The automated system reduced processing time from 2.15 to 1.28 person-minutes per fish. Mean Atlantic herring length measurements made on electronic and hardwood boards differed by 0.6 mm, and measurements made by the two systems had similar standard deviations. Automated data capture systems increase the efficiency and accuracy of fish processing and have broad applicastions in fishery assessment programs.
Sequential population analysis, currently used in many fish stock assessments, requires large sampling programs to generate catch-at-age matrices for specific seasons, areas, and gears (Gavaris and Gavaris 1983). Processing large numbers of fish for length, weight, sex, gonad weight, maturity stage, and age is often a major component of assessment workloads. Electronic devices that automate length measurements (Newton 1984; Rubec and Planck 1984; Rubec et al. 1985; Armstrong et al. 1986; Chaput et al. 1992) and data entry (Grainger and McLoughlin 1986; Armstrong et al. 1989; Morizur et al. 1992) reduce this workload by increasing the processing staff's efficiency.
Both traditional and electronic means of fish processing are prone to measurement and data entry errors. This paper describes an integrated data capture system in which a microcomputer conducts real-time plausibility checks on fish processing data generated by an balance and measuring board. Because outliers are brought to the operator's attention immediately, errors can be corrected before the fish is discarded. The system was designed for use with commercial and research samples of Atlantic herring Clupea har engus from the Gulf of St. Lawrence. To evaluate the accuracy and efficiency of this system, we report (1) the type and frequency of errors detected by error-checking routines, (2) comparative processing times for the integrated system and the traditional manual method, and (3) the repeatability of electronically recorded fish length measurements and their difference from those made on a hardwood board.
The data capture system consists of a Limnoterra FMB IV electronic fish measuring board (FMB) (Limnoterra Atlantic Inc., Kitchener, Ontario), an Ohaus Galaxy 4000 digital electronic balance (Ohaus Corp., Florham Park, New Jersey), and a laptop computer, all linked through a custom-built bi-directional interface (see Figure 1). The system is controlled by software operating on the computer and in the internal processor of the FMB.
The FMB, which is shaped like a conventional hardwood board, measures fish length by sensing the position of a magnet placed by the operator over the fish's tail. The disk-shaped magnet, attached for convenience to dissecting scissors, is 13 mm in diameter and has white stripes on its sides to facilitate alignment. Because the FMB's magnetic sensors are 2.5 mm apart, several are triggered in each measurement; the board's internal processor uses an averaging procedure to determine length to the nearest 0.5mm. The board also includes alphanumeric sensor points (A-Z, 0-9) and several other sensors for specialized instructions or data entry. The headboard houses a liquid crystal display screen on which prompts and data appear.
In the present system, the FMB is used to measure fish length and serves as an input device for entering sex and maturity data and for editing data that have failed real-time error checking.
Figure 1.- Configuration of automated data capture hardware
The top-loading Ohaus Galaxy 4000 balance transmits weights (to the nearest 0.1 g) to the computer and receives prompts and other messages from the computer via a bi-directional RS-232 port. Data transmission and taring can be initiated by the host computer or by pressing keys on the balance. Because the balance's light-emitting diode screen is large and easily read, it is used as the main display for operator prompts and warnings.
The data capture system operates on any microcomputer that uses the MS-DOS operating system and that has a RS-232 serial port. The computers used during this study were a Sharp PC-5000 and a Zenith Z-171.
The data capture system is controlled by the BASIC program CAPTHER, which manages information flow among components and provides audio and visual prompts to the operator on the balance, FMB, and computer displays. The program begins by prompting the user to input header data (sample number, area, date, etc.) for subsequent echoing to data lines (Figure 2). It then asks whether the user wants verbatim data printout during processing. Header data are then checked against maximum-minimum plausibility tables. If the data pass, the program proceeds to fish processing; otherwise the operator is prompted for corrections.
Processing begins with fish length, which is measured when a magnet is placed over the end of the tail. Fish weight is recorded by pressing a key on the balance. Gonads are then removed from the fish and weighed. Sex and gonad maturity stage are entered through the board's magnetic keypad. During these procedures the displays on the balance and the measuring board show current measurement values and prompt the operator for the next data entry. The computer screen displays the data logged for the current fish.
When all the data for one fish have been received, the computer adds a sequential fish number, rounds lengths to the nearest millimeter, and performs plausibility checks. Plausibility is checked by function limits that would reject 2-5% of verified historical data on Atlantic herring from the southern Gulf of St. Lawrence.
Figure 2.- A flowchart for the CAPTHER data capture and verification program.
The weight-length relation is flagged if either (1) log10 (weight) > 2.9 log10 (length) - 4.831 or (2) log10 (weight) < 2.9 log10 (length) - 5.629, where weight is in kilograms and length is in centimeters. These equations are derived from a generalized length-weight relation for southern Gulf of St. Lawrence Atlantic herring (log10 [weight] = 2.9 log10 [length] - 5.23; C. LeBlanc, Canada Department of Fisheries and Oceans, personal communication). If the length-weight relation fails the plausibility test, the operators prompted to either confirm or reenter the data. Revised data are retested for plausibility, and the operator has an additional chance to confirm or reenter if they fail.
Table 1.-Minimum and maximum gonad weights for southern Gulf of St. Lawrence Atlantic herring accepted by the CAPTHER program's error-checking routine.
|
MATURITY |
MINIMUM |
MAXIMUM |
|
1 |
0.0 |
0.2 |
|
2 |
0.2 |
1.5 |
|
3 |
1.0 |
20.0 |
|
4 |
10.0 |
100.0 |
|
5 |
40.0 |
120.0 |
|
6 |
10.0 |
190.0 |
|
7 |
6.0 |
20.0 |
|
8 |
1.5 |
10.0 |
The program then tests maturity stages against gonad weights and alerts the operator if weights for a given stage lie outside the limits given in table 1, which are based on southern Gulf of St. Lawrence Atlantic herring historical data (C. MacDougall, Canada Department of Fisheries and Oceans, personal communication). In case of failure the operator is prompted for confirmation or reentry as in the length-weight test. Once the plausibility tests are satisfied or overridden, header information is added and the data line is written to the file (and to the printer, if this option was preselected.) In this manner, fish are processed sequentially until the sample is completed.
Incidence and type of suspected errors flagged by CAPTHER were recorded during the routine processing of 419 Atlantic herring ( 12 samples) from commercial catches in the southern Gulf of St. Lawrence.
The processing efficiency of the automated data capture system was compared with that of the traditional manual method by measuring the time required to process 10 randomly selected Atlantic herring samples by each procedure (318 fish on the electronic board and 325 fish on the hardwood board). Two people were used to process the fish in each trial. In the manual trial, the processor took measurements, removed otoliths for later aging and dictated data to an assistant, who recorded them on paper. In the automated system test, the processor took measurements and the assistant removed otoliths. Data derived from the manual trial were entered into the computer by a trained keyboarder.
Table 2.- Incidence of errors flagged by the CAPTHER program's error-checking routine during the processing of 419 Atlantic herring.
|
NUMBER AND PERCENT OF FLAGGED ERRORS |
||||
|
TYPE OF ERROR |
NUMBER AND PERCENT OF ERRORS FLAGGED |
ACCEPTED BY OPERATOR |
PASSING AFTER CORRECTION |
FAILING AFTER CORRECTION |
|
LENGTH-WEIGHT |
2(0.5%) |
1(50%) |
1(0%) |
0(50%) |
|
MATURITY-GONAD WEIGHT |
26(6.2%) |
17(65%) |
8(31%) |
1(4%) |
|
TOTAL |
28(6.7%) |
18(65%) |
9(32%) |
1(4%) |
The repeatabalility and accuracy of FMB length measurements were tested by measuring 125 Atlantic herring seven times on a hardwood measuring board and seven times on the FMB. Samples were frozen after they were collected from commercial fisheries and thawed in cold, flowing water prior to measurement, which is the standard processing procedure. The hardwood board used in this study had a hardwood ruler embedded in a measuring surface, the sides of which sloped toward the center. The ruler was flush against the headboard, with no offset. Lengths were read to the nearest millimeter by an experienced herring processor and recorded on paper by an assistant. Fish lengths were measured on the FMB in the manner described above, but lengths, recorded by the FMB's internal algorithms to the nearest 0.5 mm, were stored in the computer by means of data received from the FMB. These lengths were later rounded to the nearest millimeter by adding 0.5 mm and taking the integer value, which is the procedure used by CAPTHER.
The CAPTHER program flagged suspected errors in 28 of 419 (6.7%) processed Atlantic herring (Table2). Of these, 18 (4.3%) were confirmed by the operator to be correct as entered. Data for the remaining 10 fish (2.4%) were revised. Nine of these passed the second error check, and data for the single fish that failed were confirmed by the operator.
Most (26 of 28, 93%) plausibility failures occurred because gonad weight was out of range for the specified maturity stage. Seventeen fish with flagged maturity-gonad weight relations were accepted as entered and nine were corrected. Only one fish failed the maturity-gonad weight test after correction. Of the two fish that produced length-weight warnings, one was accepted as entered and one was corrected. The corrected data then passed the plausibility test.
Table 3.-Processing efficiencies of manual and electronic data capture systems for processing Atlantic herring.
|
VARIABLE 3 |
||
|
|
MANUAL SYSTEM |
ELECTRONIC SYSTEM |
|
NUMBER OF SAMPLES |
10 |
10 |
|
NUMBER OF FISH |
325 |
318 |
|
PROCESSINGTIME |
658 |
410 |
|
KEYBOARDING TIME |
22 |
0 |
|
VERIFICATION TIME |
20 |
0 |
|
TOTAL PROCESSING TIME |
700 |
410 |
|
MEAN TIME PER FISH |
2.15 |
1.28 |
|
* All times are expressed in person-minutes |
||
The laboratory processing team took 658 person-minutes to process 325 Atlantic herring by the manual method (2.03 person-minutes per fish), compared with 410 person-minutes for 318 fish with the automated system (1.28 person-minutes per fish) (Table 3). Data sheets from manual processing required an additional 22 person-minutes to keyboard, and it took 20 person-minutes to verify data entry, for a total of 700 person-minutes. Total processing time for the manual method, including keyboarding and verification, was 2.15 person-minutes per fish.
Comparison of Manually and Electronically Recorded Length Measurements
Means of seven replicate length measurements 125 Atlantic herring were 306.7 mm(SD=24.9, N=875) for the hardwood board and 306.1mm (SD = 24.9, N = 875) for the electronic FMB (difference=0.6 mm, P> 0.05). For individual fish, the difference between mean manual and meanelectronic measurements averaged 1.2 mm (SD=0.9, maximum=3.9 mm; Figure 3). The two techniques produced similar within-fish mean Sds (1.2) and coefficients of variation (SD/mean=0.14). In two cases, maximum within-fish range was 31.0mm on the hardwood board, which were clearly observational error. The maximum range for an individual fish was 10.0 mm for the electronic FMB. Differences between mean hardwood and mean FMB measurements were uncorrelated with fish length as measured on the hardwood board (r=0.03, P>0.05).
Figure 3. - Frequency distribution of differences between lengths of 125 Atlantic herring measured on hardwood and electronic boards. The difference for each fish was derived by subtracting the mean of seven electronic measurements from the mean of seven hardwood measurements.
The automated data capture system described in this paper is more efficient than manual fish processing and provides more reliable data than systems in which electronic measuring boards are used as stand-alone recorders (Armstrong 1976; Newton 1984; Rubec et al. 1985; Armstrong et al. 1986). The system's data verification routine detected true errors in the measurements of 10 of 419 (2.4%) fish and allowed them to be corrected before the fish was discarded. Without real-time verification, these errors would have entered data files. The system permits aberrant values to be double-checked and confirmed, which eliminates doubts about the validity of outliers in subsequent data files. The direct input of data to computer disk also avoids the possibility of keyboarding error.
The productivity of processing staff was 68% higher with the automated system than with the manual method, largely because direct data entry freed the assistant for otolith extraction. The difference in productivity (0.87 person-minutes per fish) translates to an annual savings of 145 personnel hours for a laboratory that processes 10,000 fish per year.
In this study the electronic FMB produced a measurement average 0.6 mm smaller than that from the hardwood board. Between the FMB and the hardwood board, the mean difference of mean replicated within-fish measurements was 1.2mm. This difference has little effect because Atlantic herring are grouped by 5-mm intervals for assessment purposes. If needed, FMB results can be readily corrected by addition of a constant. Measurements from the two instruments had similar Sds and similar coefficients of variation, and tests on other units of the same model showed that data produced by the electronic FMB are similar imprecision and interobserver variability to those derived from hardwood boards (Chaput et al. 1992).
The automation of laboratory fish processing seems destined to increase as the need for timely, accurate, and cost-efficient fisheries assessments intensifies. Because the data capture and verification program is written in interpreted BASIC, an easily modified and widely applied programming language, it can be readily revised for more sensitive error checking (eg., use of a gonadosomatic index instead of a table of weights), dockside processing (eg., automatic keep-reject calls for length-stratified sampling), and applications to other species.
We thank Ginette Comeau, Colin MacDougall, and Brigette Sheehan for their assistance in the laboratory; Martina Poirier for keyboarding and verifying data; and Gerald Chaput and Dr. R. Jon Planck for critical comments. We are indebted to Paulette Hache and Ray Burkholder of Limnoterra Atlantic Inc. for sharing their insights into electronic fish measurement and the CAPTHER program.
Hit back on your browser to return to the Fish Measuring Board homepage