Basic tests of Chinese cabbage yield monitoring sensors for small-sized cabbage harvesters

Research Article
Milon Chowdhury1,2Ye-Seul Lee3Bo-Eun Jang1Yong-Joo Kim1,2Sun-Ok Chung1,2*

Abstract

Real-time yield monitoring is a critical but effective process for the assessment of the spatial variability of yield in fields, and profit information instantly. Since a few researches were conducted for upland crops yield monitoring, especially for Chinese cabbage, therefore the objective of this study was to conduct the basic tests for evaluating the potentials of mass and volume-based sensing approaches for Chinese cabbage yield monitoring for small-sized cabbage harvesters. Basic tests were conducted under laboratory conditions. Two types of sensors were used: load cells (for mass-based sensing), and CCD cameras (for volume-based sensing). For mass-based yield monitoring, an impact plate was fabricated using load cells, and installed in such a way that the cabbages reached to the collection section touching the impact plate. Mass was calculated from the load cell signals, and the effects of different dropping heights on the impact plate were also investigated. For volume-based yield monitoring, the top and side images of the cabbages were captured using CCD cameras, and volume was determined through image processing techniques. The weight of cabbage and the number of harvested cabbage were also determined. Linear calibrations with R2 of 0.97 and 0.94 were found for mass-based, and volume-based yield monitoring approaches, respectively. No significant differences were found for different cabbage falling heights. The average percentage of error for cabbage weighting and counting were 10.73% and 10%, respectively. The results showed the potentials of the candidate sensors for Chinese cabbage yield monitoring, however, further study would be necessary to minimize the effects of the factors affect the real-time yield monitoring and mapping during crop harvest.

Keyword



Introduction

Chinese cabbage is one of the most versatile and commercially valuable vegetables due to its short growth period, various use, health-promoting components, and disease prevention capability (Manchali et al., 2012; Seong et al., 2016; Shim et al., 2018). The cultivation rate of Chinese cabbage is increasing, and the usage of the Chinese cabbage harvester is also promising to cover the increasing cultivated lands and to overcome the agricultural workforce reduction due to aging, and the continuous rise of labour wages (Yu et al., 2015; Ali et al., 2019).

Yield monitoring systems have become an essential part of modern harvesters as it can provide information on the spatial variability of crop yield in the field, and gross profit instantly (Magalhães and Cerri, 2007; Maja and Ehsani, 2010; Chung et al., 2016; Kabir et al., 2018). It was not only being used in the large crop harvesters of America, Europe, or Australia but also attracted interest in the Asian countries, where harvesters and fields are relatively small. Yield information and mapping would be more effective for the small farmers for farm management practices, which would increase the yield as well as farm profit, and minimize the effects on the environment. Although yield monitoring systems for grains and other crops are quite popular, very few yield-monitoring systems are commercially available for upland vegetables, especially for Chinese cabbage.

Mass and volume-based yield monitoring approaches are commonly used in different crop harvesters, however, very limited researches were conducted for monitoring the Chinese cabbage yield using either of these approaches. Campbell et al. (1994) developed a mass-based yield monitoring method for tuber crops, where weighing cells were placed in the conveyor belt. Load cells, impact-based sensors, and bounce plates have been used in several studies for sensing yields (Demmel and Auernhammer, 1999; Ehlert, 2000; Vellidis et al., 2001; Tokunaga and Shoji, 2006; Maughan et al., 2012). Load cells are preferred as they can work accurately in various working conditions. Besides, impact-based yield sensors can measure the individual weight of crops (Fravel et al., 2013). Attachment of cushion in these sensors could reduce the error percentage (Qarallah et al., 2008). A mass-flow sensor-based peanut yield monitoring approach was developed by Thomasson et al. (2006) and found a strong correlation with the harvested load weight with R2 from 0.89 to 0.96. Volume-based yield monitoring approaches are also quite popular (Pan et al., 2007). A machine vision system on the conveyor of a harvester was attached by Hofstee and Molema (2002, 2003) to determine the volume of individual tuber crops. Gogineni et al. (2002) monitored sweet potato yield using a CCD camera and also found a high correlation (R2: 0.96) with actual weights. Image-based yield monitoring approaches were also evaluated on laboratory conditions by Persson et al. (2004) for estimating the weight of the potatoes.

Several factors affect the yield monitoring systems. Proper sensor selection, installation, calibration, and operation are major factors for accurate and precise yield estimation and mapping. Proper mounting place section is also a critical but vital issue as the sensors would face different working conditions, such as machine vibration, field slop (Zhou et al., 2014; Kabir et al., 2018). Therefore, the basic tests (hardware and software) of the yield monitoring approaches under laboratory conditions were highly necessary.

The mechanization rate for upland crop harvest is creasing in the Asian countries, especially for small fields and farms. As Chinese cabbage is one of the major vegetables, and its cultivation rate is increasing, mechanical harvesting of Chinese cabbage is also increasing. Real-time cabbage yield monitoring would be useful for instant determination of the spatial variability of cabbage in fields, and gross profit estimation. Since a few researches were conducted for Chinese cabbage yield monitoring, the objective of this study was to conduct the basic tests for evaluating the potentials of mass and volume-based sensing approaches for Chinese cabbage yield monitoring for small-sized cabbage harvesters.

Materials and methods

Chinese cabbage yield monitoring system

The self-propelled small-size (20 kW engine) cabbages harvester was considered for this study, which consisted of three work major parts (root cutting part, transportation part, and discharge part) as shown in Fig. 1. The harvested cabbage was transferred to the collection part by the transfer belt, and the tension of the transfer belt was maintained by the tension springs so that the cabbages are being transferred without dropping from the belt.

Fig. 2 shows the schematic diagram of the Chinese cabbage yield monitoring system. The major component of the mass-based yield estimation system was an impact plate, which installed in such a way that the cabbages discharged from the transportation part could be contacted before they fall to the collection part. The CCD cameras were the major component of the volume-based yield estimation system, which was placed at the end of the conveyer to capture the top and side images of the Chinese cabbages. Load cells and cameras were connected to the data logger through data acquisition modules. A LabVIEW coded program was used to collect and save the data. The detailed specification of the used components for the basic tests was mentioned in Table 1.

http://dam.zipot.com:8080/sites/pastj/images/PASTJ_20-025_image/Figure_PASTJ_20-025_F1.png

Fig. 1. Major components of the considered Chinese cabbage harvester.

http://dam.zipot.com:8080/sites/pastj/images/PASTJ_20-025_image/Figure_PASTJ_20-025_F2.png

Fig. 2. Schematic diagram of the mass and volume-based Chinese cabbage yield monitoring system.

Table 1. Specification of the components used in the Chinese cabbage yield monitoring system.http://dam.zipot.com:8080/sites/pastj/images/PASTJ_20-025_image/Table_PASTJ_20-025_T1.png

Mass-based yield monitoring

In this approach, the load cell (OBU-10, Bongshin, Korea) was used to measure the weight of Chinese cabbages. The OBU-10 was a single-point type high-precision load cell, which provided around 1/5000 accuracy of the full scale, and it was able to measure loads in various field conditions. The impact plate was composed of an acrylic plate with dimensions of 580 mm × 280 mm × 5 mm. The acrylic plate was attached with four load cells. A polyurethane cushion of 10 mm thickness was attached with the acrylic plate to absorb the shocks. The weight measuring capacity of each selected load cell was 10 kg, where the average weight of Chinese cabbage was around 3 kg to 5 kg. The distance between the conveyer belt and the impact plate was adjustable to evaluate the effects of the falling height on yield monitoring (Fig. 3). The time difference between two cabbages dropped into the impact plate, and data acquisition intervals were adjusted and kept the same for saving only one averaged data (weight) for each cabbage. Finally, the number of harvested cabbage was determined by counting the total number of data.

http://dam.zipot.com:8080/sites/pastj/images/PASTJ_20-025_image/Figure_PASTJ_20-025_F3.png

Fig. 3. Mass-based yield monitoring: (A) schematic diagram of the mass-based Chinese cabbage yield monitoring system, and (B) photo of the fabricated impact plate.

Volume-based yield monitoring

The volume-based yield monitoring approach depends on the images captured by the CCD cameras. In the laboratory condition, top and side images of the Chinese cabbages were captured using CCD cameras with a resolution of 1288 horizontal × 960 vertical (progressive scanning, and continuous video outputs). Two 35 W halogen lamps were installed. Considering the light condition, 1/50 s shutter rate was selected, and the lens aperture was adjusted. Camera utility software (GAPI SDK v2.6, Baumer Int., Stockach, Germany) was installed in the computer for capturing images (Fig. 4). The cabbage to cabbage distance was 30 cm. External light and internal reflectance were blocked using a rectangular Conveyor belt with black colored paper, which would be mounted at the end of the conveyer belt of the cabbage harvester during the field test.

The weight of cabbage was calculated by multiplying the volume and density. A mathematical formula (Equation 1) was used to determine the volume of Chinese cabbage from two-dimensional filtered images captured by the CCD cameras. The shape of cabbage was considered as ellipsoidal, where the major axis (the longest shape captured by the top CCD camera) and minor axis (round shape captured by the side camera) were represented the height, and length and width, respectively.

http://dam.zipot.com:8080/sites/pastj/images/PASTJ_20-025_image/PASTJ_20-025_eq1.png (1)

where, V: volume of Chinese cabbage (m3), a: major axis (m), and b, c: minor axis (m). The density of Chinese cabbage was determined by dividing the mass by volume, where the mass was measured by an electric balance, and actual volume was measured by placing the cabbage in a cylinder filled with certain amount of water. The mass coefficient was determined from the ratio of the estimated mass and actual mass. Equation 2 represents the mass estimation formula.

http://dam.zipot.com:8080/sites/pastj/images/PASTJ_20-025_image/PASTJ_20-025_eq2.png (2)

where, M: weight of Chinese cabbage (kg), Cm: mass coefficient, ρ: density of Chinese cabbage (kg/m3), V: volume of Chinese cabbage (m3).

http://dam.zipot.com:8080/sites/pastj/images/PASTJ_20-025_image/Figure_PASTJ_20-025_F4.png

Fig. 4. Volume-based yield monitoring: (A) schematic diagram of the volume-based Chinese cabbage yield monitoring system, and (B) image acquisition of Chinese cabbage in the laboratory condition.

Cabbage counting

A flow chart algorithm (Fig. 5) was developed for counting the number of harvested cabbage, as mentioned by Lee et al. (2018). At first, the system objects were formed to detect the video frames, where the cabbage images were captured. An array of tracks was initiated then, where each track represented a moving cabbage of the video. After detecting the cabbage through image processing, the required information, (i.e., centroid coordinate, bounding box, major, and minor axis dimension) were collected. In the next steps, the centroid of each track in the current frame was determined, and the bounding box was improved accordingly. Then, the cost was calculated. The cost value less than 200 indicated the successful identification of cabbage from the track, and the age of the cabbage was increased by one. However, the track was deleted when the cost value was over 200, as it was predicted too damaged. The process returned to the ‘predict new location of existing track’ step until the age of the cabbage became 200. After perfect matching, the number of cabbage was counted and the process was stopped.

http://dam.zipot.com:8080/sites/pastj/images/PASTJ_20-025_image/Figure_PASTJ_20-025_F5.png

Fig. 5. Flow chart of the Chinese cabbage counting algorithm.

Calibration and Data Analysis

Several factors affect the accuracy of the mass and volume-based yield monitoring system. The position of load cells, the thickness of the cushion, cabbage falling height, the weight of cabbage, and attachment of the impact plate could affect the mass-based yield monitoring approach. First, the output signals of the load cells (impact plate) were checked for different known weights (1kg, 5 kg, and 10 kg). Then the actual weight of 10 Chinese cabbages was measured and compared with the load cell output values for calibration.

To evaluate the effects of the falling height on yield monitoring, 5 different distance (10, 20, 30, 40, and 50 cm) were selected and tested. Three replications were performed for each test. The data acquisition rate from the load cells was 1 kHz. The coefficient of determination (R2) was calculated for every basic test to express the variability between the considered factors.

Images of the Chinese cabbages were captured and processed for the volume-based yield analysis. A Contrast-limited Adaptive Histogram Equalization (CLAHE) was applied to the captured images and turned into binary images by setting the threshold to 80. Then the images were dilated with a vertical line structuring element, and the number of pixels was counted from the overlying white area whose value was 1 in the processed image. The MATLAB software (vR2013b (8.2.0.701), MathWorks Ins., Natick, Massachusetts, USA) was used for this whole process. Finally, the weight of each cabbage was determined by multiplying the volume and specific density. The calculated weight and the actual weight of each Chinese cabbage were also compared during the calibration.

Results and discussion

Calibration of mass and volume-based yield monitoring approaches

Fig. 6(a) shows the comparison of Chinese cabbage weight and load cell output for the calibration of the mass-based yield monitoring approach. The coefficient of determination (R2) value was 0.97 with a coefficient of interception of 0.0047 and a gradient of 0.0006. The root means square error (RMSE) value was 0.0013 V. It indicated that measurement of Chinese cabbage yield would be recommended using the impact plate with load cells. The comparison of actual weight and calibrated weight of Chinese cabbages for the calibration of the volume-based yield monitoring approach showed in Fig. 6(b). The coefficient of determination (R2) value was 0.94 with a coefficient of interception of 1.03 and a gradient of 0.33, and the root mean square error (RMSE) value was 0.33 kg, which also indicated the good possibility of Chinese cabbage yield monitoring using the CCD cameras.

http://dam.zipot.com:8080/sites/pastj/images/PASTJ_20-025_image/Figure_PASTJ_20-025_F6.png

Fig. 6. Calibration: (A) comparison of Chinese cabbage weight and load cell output for the calibration of the mass-based yield monitoring, and (B) comparison of Chinese cabbage weight and calibrated weight for the calibration of the volume-based yield monitoring.

Mass-based yield monitoring

The total number of saved data indicated the number of harvested cabbages, and the weight of each cabbage was calculated from the average output signal of the load cells. The counting and weighting errors were 10% and 7.53%, respectively. Fig. 7 shows the effect of different falling distances (10 cm, 20 cm, 30 cm, 40 cm, and 50 cm) of Chinese cabbages. However, a clear difference was observed for different falling height, but no significant variation wasfound in these tests. The coefficient of determination (R2) varied from 0.97 to 0.98 with a negligible variation of coefficient of interception and gradient. The root means square error (RMSE) also varied from 0.12 to 0.13 V. The effects of falling distance more than 50 cm, and the polyurethane cushion thickness of more than 10 mm also need to be investigated. Mass-based yield monitoring is a common approach, and several researchers found good correlation results during the tests, such as Kabir et al. (2018) applied this approach for potato yield monitoring and found a strong correlation with R2 values ranged from 0.96 to 0.98. Similarly, Thomasson et al. (2006) also observed R2 values from 0.89 to 0.96 during peanut yield monitoring using the mass-flow sensor.

Volume-based yield monitoring

The pixel size of an image was 3.75 × 3.75 μm. The area was calculated by counting the pixel, where 559 pixels indicated 1 cm2 reference area. The volume and weight of Chinese cabbage were determined by multiplying the area with height, and volume with density, respectively. Fig. 8(a) shows the steps of image processing, and Fig. 8(b) shows the volume calculation. The performance of the volume-based yield monitoring system might be improved by capturing good quality images using the LED light and RGB camera instead of halogen light and CCD camera. Besides this, the vibration level needs to be minimized during the field test for preventing the degradation of image quality (Xu et al., 2003). The volume-based yield monitoring approach was considered as an alternative of mass-based yield monitoring. Lee et al. (2018) evaluated a vision-based potato yield detection and counting system with a minimum average percentage error (around 7%). The image-based approach also used in different studies as a potential sensing method for yield monitoring of crops (Gogineni et al., 2002; Hofstee and Molema, 2002, 2003; Persson et al., 2004; Dunn et al., 2006).

The calculated mass coefficient and density of the Chinese cabbage were 0.786 and 0.3212 g/cm3, respectively. The average percentages of error for cabbage weighting and counting were 13.93% and 10%, respectively. The lower percentage of errors could be occurred due to the bigger size and weight of Chinese cabbage. However, the low number of samples, and slow cabbage harvesting rate could also affect the error percentage.

Performance comparison between mass and volume-based approach

The performance of the mass and volume-based yield monitoring approaches was compared based on the calibration results and value of the coefficient of determination (R2). Both approaches showed a strong correlation with the measured weight (R2: 0.97 for mass-based, and R2: 0.94 for volume-based approach), and the minimum root means square error. However, the volume or image-based yield monitoring approach showed a lower performance compared to the mass-based yield monitoring approach. Kabir et al. (2018) also drew a similar conclusion during the potato yield monitoring. Several studies reported high error percentages during the yield monitoring of small-sized crop. Large volume and weight of Chinese cabbage might be a reason for the good results in this study. Based on the affecting factors, such as soil condition, field slop, vibration, and speed of the harvester, yield monitoring results could be varied.

http://dam.zipot.com:8080/sites/pastj/images/PASTJ_20-025_image/Figure_PASTJ_20-025_F7.png

Fig. 7. Evaluation of the effect of different falling distances of Chinese cabbages: (A) distance 10 cm, (B) distance 20 cm, (C) distance 30 cm, (D) distance 40 cm, and (E) distance 50 cm.

http://dam.zipot.com:8080/sites/pastj/images/PASTJ_20-025_image/Figure_PASTJ_20-025_F8.png

Fig. 8. Volume-based Chinese cabbage yield monitoring approach: (A) image processing steps (a), and (B) volume calculation of cabbage.

Conclusion

This study was conducted to evaluate the potentials of mass and volume-based sensing approaches for the Chinese cabbage yield monitoring for small-sized cabbage harvesters. For this purpose, the basic tests of these approaches were carried out under laboratory conditions. An impact plate attached with load cells was fabricated and tested to measure the weight of Chinese cabbages. The effects of falling height on the weight measurement were also evaluated to find out the proper installation position of the impact plate. CCD cameras were used to capture the top and side views of Chinese cabbage. The volume, weight, and the number of harvested cabbage were calculated through image processing techniques. A strong correlation with R2 values 0.97 and 0.94 were observed for mass, and volume-based yield monitoring approaches, respectively. The falling heights (up to 50 cm) didn’t show any significant impact in weight measurement. The average percentages of error for cabbage weighting and counting were 10.73% and 10%, respectively. The results indicated the potentials of both approaches for the yield monitoring of Chinese cabbage and similar type upland crops., however, further research are necessary to evaluate the effects of sample numbers, soil condition, field slop, vibration, and speed of the harvester during on-site tests and yield mapping.

Acknowledgments

This work was supported by Korea Institute of Planning and Evaluation for Technology in Food, Agriculture and Forestry (IPET) through Agriculture, Food and Rural Affairs Convergence Technologies Program for Educating Creative Global Leader, funded by Ministry of Agriculture, Food and Rural Affairs (MAFRA) (Project No. 320001-4), Republic of Korea.

References

1  Ali M, Lee YS, Kabir MSN, Kang TK, Lee SH, Chung SO. 2019. Kinematic analysis for design of the transportation part of a tractor-mounted Chinese cabbage collector. Journal of Biosystems Engineering 44(4):226-235. 

2  Campbell RH, Rawlins SL, Han SF. 1994. Monitoring methods for potato yield mapping. In American Society of Agricultural Engineers. Meeting (USA). 

3  Chung SO, Choi MC, Lee KH, Kim YJ, Hong SJ, Li M. 2016. Sensing technologies for grain crop yield monitoring systems: A review. Journal of Biosystems Engineering 41 (4):408-417. 

4  Demmel M, Auernhammer. 1999. Local yield measurement in a potato harvester and overall yield patterns in a cereal-potato crop rotation. American Society of Agricultural and Biological Engineers 99:11.  

5  Dunn M, Billingsley J, Bell D. 2006. Vision based macadamia yield assessment. Sensor Review 26 (4):312-317. 

6  Ehlert, D. 2000. Measuring mass flow by bounce plate for yield mapping of potatoes. Precision Agriculture 2(2):119–130. 

7  Fravel JB, Kirk KR, Monfort WS, Thomas JS, Henderson WG, Massey HF, Chastain JP. 2013. Development and testing of an impact plate yield monitor for peanuts. American Society of Agricultural and Biological Engineers, Paper number: 131620969.  

8  Gogineni S, Thomasson JA, Wooten JR, White JG, Thompson PR, Shankle M. 2002. Image-based sweetpotato yield and grade monitor. American Society of Agricultural and Biological Engineers, Paper number: 021169. 

9  Hofstee JW, Molema GJ. 2002. Machine vision based yield mapping of potatoes. American Society of Agricultural and Biological Engineers, Paper number: 021200 . 

10  Hofstee JW, Molema GJ. 2003. Volume estimation of potatoes partly covered with dirt tare. American Society of Agricultural and Biological Engineers, Paper number: 031001 . 

11  Kabir M, Myat Swe K, Kim YJ, Chung SO, Jeong DU, Lee SH. 2018. Sensor comparison for yield monitoring systems of small-sized potato harvesters. In 14th International conference on precision agriculture . 

12  Lee YJ, Kim KD, Lee HS, Shin BS. 2018. Vision-based potato detection and counting system for yield monitoring. Journal of Biosystems Engineering 43(2):103-109. 

13  Magalhães PSG, Cerri DGP. 2007. Yield monitoring of sugar cane. Biosystems Engineering , 96 (1):1-6. 

14  Maja JM, Ehsani R. 2010. Development of a yield monitoring system for citrus mechanical harvesting machines. Precision agriculture 11 (5):475-487. 

15  Manchali S, Murthy KNC, Patil BS. 2012. Crucial facts about health benefits of popular cruciferous vegetables. Journal of Functional Foods, 4(1):94-106. 

16  Maughan JD, Mathanker SK, Grift TE, Hansen AC. 2012. Yield monitoring and mapping systems for hay and forage harvesting: a review. American Society of Agricultural and Biological Engineers, Paper number: 121338184. 

17  Pan G, Li FM, Sun GJ. 2007. Digital camera based measurement of crop cover for wheat yield prediction. Geoscience and Remote Sensing Symposium. IEEE 797-800. 

18  Persson DA, Eklundh L, Algerbo PA. 2004. Evaluation of an optical sensor for tuber yield monitoring. Transactions of the ASAE 47(5):1851-1856. 

19  Qarallah B, Shoji K, Kawamura T. 2008. Development of a yield sensor for measuring individual weights of onion bulbs. Biosystems Engineering 100:511- 515. 

20  Seong GU, Hwang IW, Chung SK. 2016. Antioxidant capacities and polyphenolics of Chinese cabbage (Brassica rapa L. Ssp. Pekinensis) leaves. Food Chemistry 199:612-618. 

21  Shim JY, Kim HY, Kim DG, Lee YS, Chung SO, Lee, WH. 2018. Optimizing growth conditions for glucosinolate production in Chinese cabbage. Horticulture, Environment, and Biotechnology 59 (5):649-657. 

22  Thomasson JA, Sui R, Wright GC, Robson AJ. 2006. Optical peanut yield monitor: development and testing. Applied Engineering in Agriculture 22(6):809-818. 

23  Tokunaga J, Shoji K. 2006. Development of potato yield sensor to measure the mass of individual tubers. Preprints of Third IFAC International Workshop on Bio-Robotics, Information Technology and Intelligent Control for Bioproduction Systems (Bio-Robotics III), pp 239-243, Sapporo, Japan. 

24  Vellidis G, Perry CD, Durrence JS, Thomas DL, Hill RW, Kvien CK, Hamrita TK, Rains G. 2001. The peanut yield monitoring system. Transactions of the ASAE 44 (4):775-786. 

25  Xu P, Hao Q, Huang C, Wang Y. 2003. Degradation of modulation transfer function in push-broom camera caused by mechanical vibration. Optics & laser technology, 35(7):547-552. 

26  Yu SC, Shin SY, Kang CH, Kim BG, Kim JO. 2015. Current status of agricultural mechanization in South Korea. In: Proceeding of ASABE Annual International Meeting (paper no. 152189653). St. Joseph. Michigan, USA. 

27 Zhou J, Cong B, Liu C. 2014. Elimination of vibration noise from an impact-type grain mass flow sensor. Precision Agriculture 15(6):627-638.