Tomato seedlings were cultivated in 128-pore plastic trays (16×8); the pores were shaped like an inversed quadrangular platform and their sizes were: upper width = 30 mm, lower width = 13 mm, height = 40 mm; included angle between two face-to-face walls = 24°; volume = 19.45 cm ³ . The seedling substrate was a mixture of sphagnum peat moss, perlite and vermiculite (3:1:1 v/v/v); the trays were filled with the substrate at the 1.2-fold volume of pores ( Kumar & Raheman, 2010 ; Nandede et al. , 2014 ). Xinken 84-7 tomato seeds were embedded at the depth of 5 mm and cultivated in a modern greenhouse at Jiangsu University for a growth cycle of 45 days.

The tomato root lumps were scanned by an X-ray µCT 100 instrument (SCANCO Medical AG, Switzerland) in the Key Laboratory of Modern Agricultural Equipment and Technology, Jiangsu University. The X-ray µCT scanner had the operating system OVDM (Open VMS­/­DEC windows Motif 1.1, designed by SCANCO Medical AG) and image processing language (IPL) that can be used to three-dimensionally reconstruct the roots and pores and to measure the volume and distance between two random points in space.

During the trials, the tomato plug seedling and the pick-up device must be put into a barrel-shaped contai­ner (diameter = 100 mm, height = 120 mm) (Fig. 1). The pick-up arm and the pick-up pin formed a gripper. Then the position-regulating hob was rotated, so two grippers could transversally contract or stretch in a symmetrical way, and the pick-up arms could be installed with pick-up pins of different shapes. Generally, pick-up pins are metal-made; in order to avoid metal artifacts during scanning ( Kalender et al. , 1987 ; Roberson et al. , 1997 ), we used pick-up pins made of glass fibers and in density of 2.08 g/cm ³ . The foundation, supporting shore and cover board of the pick-up device were made of polyvinyl chloride.

To accurately control the initial pick-up angle θ and contraction b when pick-up pins are inserted into a root lump, we made seedling boxes to hold the plug seedlings (Fig. 1). The pores in the seedling boxes for placement of root lumps were of the same sizes as the 128-pore seedling trays. The wall thickness and density of the boxes were 2 mm and 1.01 g/cm ³ , respectively.

During scanning, the µCT parameters were set as: voltage = 45 Kvp, current = 75 µa, resolution = 50 µm, integral time interval = 400 ms; the signals were denoised using an AL0.1 filter ( Zappala et al. , 2013a ). To study the pore distribution on CT graphs at different pick-up states, we set the same initial reference line for each scan. The water content of root lumps would affect the clearness of 3D root reconstruction, and Zappala et al. (2013b) found the root extraction clearness was the optimal when the soil water content in scanning ripe seedlings was 25-35%. In our study, we set the water contents of tomato root lumps to be 30-35%, at which the root lump density was 1.72 g/cm ³ .

When the water content of tomato root lumps is 30-35% (Fig. 2A), the densities of sphagnum peat moss and perlite are 1.65 and 1.57 g/cm ³ , respectively, which are very similar. The very low contrast on the CT graphs of root lumps (Fig. 2B) makes them hardly distinguishable. On the contrary, the vermiculite with higher density (2.01 g/cm ³ ) than the other two components is very bright. On the CT graphs, the main roots and pores can be clearly located, so closed boundary lines around the main roots and pores were plotted separately. The threshold of the selected zone can be read in the Open VMS/DECwindows Motif 1.1 (OVDM) system. Thresholds of the main roots and pores were determined separately from multiple CT graphs and then averaged. The final thresholds of roots and pores were 64-73 and 0-15, respectively. Then with the threshold partition command on IPL, the roots and pores were extracted and three-dimensionally reconstructed (Fig. 2C, 2D). However, the 3D root images also showed some substrate grains, but did not show a part of fine lateral roots. This was because the root and seedling substrate could not be fully differentiated due to their low contrast on the CT images ( Mooney et al. , 2012 ; Stefan et al. , 2015 ) and the scanning resolution could also affect the quality of root reconstruction ( Richard et al. , 2012 ; Saoirse et al. , 2013 ).

To validate the precision of root segmented extrac­tion, we first measured the extracted root volume to be 354.6 mm ³ using IPL, and then washed away the substrate in the lumps. The roots were put into 5-mL dosimeters, which were injected with known volumes of distilled water until the roots were all covered. Then the volumes at the liquid level were read. The root volumes were determined by subtracting the volume of distilled water from the liquid level volume as-read, and were found to be 336.4 mm ³ . The root volumes determined from µCT segmented extraction method were 5.13% larger than those measured by the dosimeter method. The surfaces of root lumps were photographed using a digital camera. The distance at two random points on the lateral roots of a lump were measured on ImageJ ( Rasband, 2011 ). Totally four segments of lateral roots were measured (Fig. 2A) and the average root length was 6.23 mm. On the 3D root images, the corresponding lateral root length was measured to be 6.02 mm on average, with error of 3.37% from the above method. Thus, the errors of the µCT segmented extraction method were very low.

To validate the precision of the pore segmented extraction method, we obtained CT images of root lump surface layer using µCT (Fig. 3A), segmented and extracted the pores (Fig. 3B). The lacunarity was measured to be 3.81%. The pore fractal dimension determined on the plug-in BoneJ of ImageJ ( Michael et al. , 2010 ) was 1.153. Then the root lump surfaces as-scanned were photographed (Fig. 3C), and the surface pores of root lumps ( Sun et al. , 2017a ) could be clearly identified. The boundary lines of the pores were plotted on ImageJ and the lacunarity was 3.93%, which was larger than that estimated from µCT (3.15%), and the pore fractal dimension was 1.151, which was not significantly different from that determined from µCT. Thus, the errors of pores determined from the µCT segmented extraction method are very small.

With pin shape, diameter and the initial pick-up angle as the test factors ( Choi et al. , 2002 ; Han et al. , 2015 ), tests were conducted to optimize the pick-up parameters that reduced the destruction of root lumps. The tests were conducted in two steps. The gripper was designed as two-finger and four-pin (Fig. 4). The depth and contraction of a pick-up pin inserted into a root lump were 35 and 5 mm, respectively. Firstly, a 2.5 mm circular pick-up pin (Fig. 4A) was used to pick up a root lump at an initial pick-up angle θ of 21°; then the roots and pores were extracted from the lump and three-dimensionally reconstructed, aiming to find out the main causes of root lump destruction. Then the effects of pin shape, pin diameter and initial pick-up angle on root lump destruction were studied, aiming to find out the pick-up parameters when the root lump destruction was low. The pick-up conditions were: pick-up pin was circular or flat (Fig. 4), pin diameter d was 2, 2.5 or 3 mm, and the initial pick-up angle θ was 18°, 21° or 24°.

Since µCT can only scan stationary substances and to study the distributions of roots and pores during the pick-up process, we considered four states of each specimen: pick-up pin was not inserted into root lump (S1), pick-up pin was inserted into root lump (S2), pick-up pin was moved to the medium contraction (S3), and pick-up pin was moved to the maximum contraction (S4).

To study the root and pore distributions during the pick-up process, we divided each root lump into five equal zones vertically and six equal zones horizontally (Fig. 5) and defined the root distribution density as the ratio of root volume in each zone to the volume of root lump in this zone. Then the root volume, root distribution density, lacunarity, and pore fractal dimension in each zone were quantitatively analyzed. In each group, 10 samples were scanned, and totally there were 100 samples. The test data were stored on Excel. Single-factor analysis of variance involving root volume, root distribution density, and lacunarity was conducted on SPSS 18.0. The mean values were compared via Tukey test, which required relatively small sample size.

Figure 6A shows the analytical values of vertical root volume V _RV . Clearly, the V _RV at each pick-up state first decreased and then increased. The statistics of V _RV are summarized in Table 1. When the pick-up pins contracted to state S4, the V _RV in zone V1 and V2 decreased, but V _RV in other zones rose. The change of V _RV varied insignificantly between -2.4 and 4.8 mm ³ ( p > 0.05). Figure 6B shows the analytical values of horizontal root volume V _RH . Clearly, the V _RH at each pick-up state first increased and then decreased, which indicates a nearly symmetrical distribution. When the pick-up pins contracted to the largest displacement, V _RH decreased significantly in zones H1 and H6 by 4.2 and 4.1 mm ³ , respectively ( p < 0.05), but rose in other zones, V _RH increased insignificantly from 3.4 to 3.9 mm ³ ( p > 0.05).

Figure 6A shows the analytical values of vertical root distribution density D _RV . Clearly, D _RV in zone V5 was significantly larger than at other zones. The statistics of D _RV are summarized in Table 1. When the pick-up pins contracted to the largest displacement, the D _RV in zone V1 and V2 decreased, but D _RV in other zones rose. The change of V _RV varied insignificantly between −0.05% and 0.14% ( p > 0.05). Figure 6B shows the analytical values of horizontal root distribution density D _RH . Clearly, D _RH in zones H1 and H6 was significant­ly larger than at other zones. When the pick-up pins contracted to the largest displacement, the D _RH in zones H1 and H6 decreased significantly by 0.42% and 0.41%, respectively ( p < 0.05), but D _RH in other zones rose insignificantly from 0.08% to 0.13% ( p > 0.05).

Figure 7 shows the 3D root images. At each pick-up state, the roots formed similar spatial structures. Specifically, the lateral roots grew divergently from the main roots to the surroundings until contacting the pore walls. Then the lateral roots changed the growing direction and went on to grow along the pore walls and wind the root lumps. The distance between the outmost lateral roots at the same position under state S1 and state S4 was measured (Fig. 7A2, 7D2). Clearly, the horizontal distances between lateral roots were 24.57 and 23.19 mm, respectively, with a decrease of 1.38 mm, while the vertical distances were 24.86 and 25.21 mm, respectively, with an increment of 0.35 mm. Han et al. (2013) used a mechanical tester to compress cucumber root lumps and concluded that the root lumps were elastic-plastic bodied wrapped by roots and the root lumps reached the yield limit and plastically deformed when the compression force was 6.33 N. When a root lump was picked up by pins, shear stress and pressure stress existed inside the root lump. Under the action of the shear stress, the raising substrate shifted in the direction perpendicular to the pick-up pin contraction, bringing the roots to move outside the root lump. Under the pressure stress, the substrate moved in the root lump in parallel to the pick-up pin contraction ( Karmakar et al. , 2007 ; Mojtaba et al. , 2013 ). When the pick-up pin contracted horizontally, the distance between the outmost lateral roots decreased horizontally and increased vertically (Fig. 7).

At the initial pick-up angle of 21º, when the pin diameter enlarged from 2 to 3 mm, the reduction of distance between the outmost lateral roots at the pin contraction direction rose from 1.27 to 1.56 mm, and the increment in the distance between the outmost lateral roots perpendicular to the pin contraction direction increased from 0.29 to 0.41 mm. At the pin diameter of 2.5 mm, when the initial pick-up angle enlarged from 18º to 24 º, the reduction of distance between the outmost lateral roots at the pin contraction direction decreased from 1.61 to 1.29 mm, and the increment in the distance between the outmost lateral roots perpendicular to the pin contraction direction declined from 0.43 to 0.31 mm. Clearly, the increments in the distance between the outmost lateral roots were very small, which was because the hole walls restricted the outward expan­sion of the root lumps. During the pick-up process, though the outmost lateral roots shifted, the spatial distribution of roots did not change significantly, as the roots always wrapped the raising substrate and thereby prevented the root lumps from loosening ( Shaw, 1993 ; Han et al. , 2015 ).

The root lumps were scanned to generate 800 CT images, from which one image every 180 images was selected. Totally, four images were selected, in which the first to third images all were passed through by pick-up pins (Fig. 8).

From the first to the third images, the pores at state S1 were scattered; as the contracting displacement of the pick-up pins was intensified, more new pores were formed and connected to form larger pores ( Keyes et al. , 2013 ). However, on the fourth image, the number of pores decreased when the contracting displacement of pick-up pins was enhanced. Table 2 shows the pore areas on the CT images at different states. The pore areas were gradually enlarged on the first to third images, but were reduced on the fourth image, while the changes of pore areas varied significantly between -4.7 and 8.7 mm ² on the four images ( p < 0.05). The areas of largest pores on the CT images were statistically analyzed on Image J (Table 2). When the pick-up pins contracted to state S4, the pore areas on images 1 to 3 increased by 1.76, 1.22 and 0.71 mm ² , respectively, but that in image 4 decreased by 0.49 mm ² ; the largest pore area varied very significantly among the four images ( p < 0.05). Clearly, the newly-formed pores mainly appeared on the CT images in contact with pick-up pins, but the pore areas decreased in the images without contact with the pins.

Figure 9A shows the analytical values of vertical lacunarity D _PV . Clearly, the D _PV at each pick-up state changed differently. The statistics of lacunarity at different states were summarized in Table 3. When the pick-up pins contracted to the largest displacement, the changes of D _PV in zones V1 to V5 varied significantly between -4.09% and 4.74% ( p < 0.05). Figure 9B shows the analytical values of horizontal lacunarity D _PH . Clearly, the V _PH was almost symmetrically distributed at each pick-up state. When the pick-up pins contracted to state S4, the changes of D _PH in zones H1 to H6 varied significantly between 0.32% and 5.94% ( p < 0.05).

Figure 9A shows the analytical data of vertical pore fractal dimension F _PV and Table 3 lists the statistical data. During the pick-up process, the F _PV first decreased and then increased in all zones from V1 to V4, but F _PV in zone V5 declined. Figure 9B displays the analytical values of horizontal pore fractal dimension F _PH . Clearly, the F _PH first declined and then increased in all zones, and the F _PH was almost symmetrically distributed in all the pick-up states.

Figure 10 shows the spatial pore distributions at different pick-up states. At state S1, the pores accumulated at the top and bottom of the root lumps, but the pores at other positions were scattered. The permeable top of root lumps facilitated water eva­poration, while the roots at the bottom promoted the migration of the raising substrate during the growing period (Fig. 7). These two situations both led to pore aggregation. At the S2 state, the pick-up pins inserted into the root lumps urged the migration of the raising substrate, while cracks and newly-formed pores gathe­red around the pick-up pins; under the action of shear stress, new pores also appeared in between the pick-up pins ( Espinoza et al. , 2016 ; Sun et al. , 2017b ). Under the pressure stress, the raising substrates far from the pick-up pins mutually extruded, leading to pore closure, and especially when the pick-up pins did not reach the bottom of root lumps, the number of pores significantly decreased ( Karmakar et al. , 2007 ; Mojtaba et al. , 2014 ). From state S3 to state S4, the pin contraction and migration directly avulsed the root lumps, and the cracks outside the pick-up pins started to expand and generated more new pores. In some areas, the cracks already expanded to the outmost end of the root lumps. Under the pressure stress, the pores in between the pick-up pins were closed, and at the exterior areas far from the pick-up pins, the pores were not obviously changed after the pick-up pins inserted into the root lumps.

In the zones V1 to V4, H2 and H5 which directly contacted the pick-up pins, the lacunarity gradually increased, but the fractal dimensions firstly decreased and then rose, and the fractal dimensions at state S4 surpassed those at state S1 (Fig. 9). At state S2, the pores and cracks were distributed nonuniformly in all zones and appeared more around the pins, but the number of pores far from the pins declined. The formation of cracks enlarged the lacunarity, while the local aggregation of cracks and the decreasing number of pores distant from the pins together led to a reduction of fractal dimensions ( Annette & Martin, 2005 ; Lars et al. , 2012 ). At states S3 and S4, the expanding cracks and the new pores spread to large space in all zones, so that the lacunarity and fractal dimensions increased in zones V1 to V4, H2 and H5 ( Zhao et al. , 2017 ). At state S2, in zones H1, H3, H4 and H6 that did not contact with the pins, the pores were closed and thus the lacunarity and fractal dimensions decreased. When the pick-up pins contracted and moved to state S4, the cracks and pores around the pins already expanded to zones H1 and H6, while the top of the pins moved to H3 and H4. The numbers and distrubtive areas of both cracks and pores all increased in the above four zones, leading to the increase of lacunarity and fractal dimensions to varying degrees. When the water contents of root lumps increase, the small pores in the root lumps are filled with water, which prevented the new pore generation and crack expansion. However, too high water contents would weaken the anti-deformation ability of root lumps, and consequently, the root lumps would fall off when the grippers pick up root lumps. As reported, when the water contents of root lumps were 50%-65%, the anti-destruction ability and pick-up success rate of root lumps were the optimal ( Mao et al. , 2014 ; Han et al. , 2015 ).

The pick-up pin to pick up a root lump should overcome the adhesion between the root lump and the hole walls, which leads to stress concentration in zone V4 (where cracks are enlarged) and in zone V5 (where the number of pores decreases) and finally breaks the root lump. To diminish the occurrence of fracture, workers should insert the pick-up pins into the root lumps as deep as possible ( Tong et al. , 2013 ; Jiang et al. , 2017 ). In zones H2 and H5 where the pick-up pins contracted, the lacunarity increased the most largely. If the pores and cracks in these two zones expanded to outside the root lumps or connected to form fracture surfaces, the raising substrate and the root lumps would be partially separated, leading to the destruction of root lumps. The destruction of root lumps can be deminished by optimizing the pin structures and pick-up parameters ( Yang et al. , 1991 ; Ryu et al. , 2001 ; Jiang et al. , 2017 ).

The distributions of roots and pores during the pick-up process show the pore aggregation and crack formation were the main causes of root lump destruction. V4 was called the fracturable zone, while H2 and H5 were the breakable zones. Root lump destruction can be evaluated using the lacunarity of these zones at state S4. During the pick-up of root lumps, the root lumps would be destroyed when the cracks expanded to a certain degree. The cracks in Figure 11 spread to the two ends of the root lumps, where destruction was observed. A rectangular frame was plotted with the longest distance of cracks as the diagonal line. Then the lacunarity of 60 consecutive CT graphs within this rectangle was statistically analyzed to be 16.8%, which was considered as the destruction index. It was thought that the fracturable zone and breakable zone in which the lacunarity reached this index would be destroyed.

When the pin diameter was 2 mm and the initial pick-up angle was 24º, the lacunarity in the fracturable zone (D _PD ) and the lacunarity at the breakable zone (D _PS ) for the circular pick-up pins both minimized. The D _PD and D _PS of flat pick-up pins both maximized when the pin diameter was 3 mm and the initial pick-up angle was 18º. All tested factors significantly affected both D _PD and D _PS ( p < 0.05) (Fig. 12).

The D _PD and D _PS of both circular and flat pins rose with the increase of pin diameter (Fig. 12A). At the pin diameter of 2, 2.5 and 3 mm, the D _PD of the circular pins was 7.1%, 8.5% and 9.8%, respectively, and the D _PS was 8.1%, 9.4% and 11.3%, respectively, while the D _PD of flat pins was 8.2%, 10.6% and 12.4%, respectively, and the D _PS was 9.0%, 10.8% and 12.9%, respectively. Clearly, the D _PS of the flat pins maximized at the pin diameter of 3 mm, which was 3.9% smaller than the destruction index, so the possibility of root lump breakage was higher. At the pin diameter of 2 mm, the D _PS and D _PD were both more than 7.8% smaller than the destruction index for both the circular pins and the flat pins, so the possibility of lump destruction was lower. Thus, to reduce the destruction of root lumps, the pick-up pin diameter should be decreased. However, to get enough clamping force in the thin pins and to overcome the adhesion between root lumps and hole walls, the shrinkage of pins should be increased ( Choi et al. , 2002 ; Tong et al. , 2014 ), which would promote crack expansion. Moreover, thin pins would be bent due to the insufficient rigidity, which was unfavorable for the pick-up of seedlings. Thus, the destruction to root lumps did not decrease with the reduction of pin thickness. For multi-pin grippers, the pins in diameter of 2 mm already had excellent pick-up performance ( Han et al. , 2015 ).

As the initial pick-up angle was enlarged, the D _PD and D _PS both gradually declined for both the circular and flat pins (Fig. 12B). At the initial pick-up angle of 18º, 21º and 24º, the D _PD of the circular pins was 9.9%, 8.5% and 7.4%, respectively, and the D _PS was 11.7%, 9.4% and 8.5%, respectively, while the D _PD of flat pins was 13.6%, 10.6% and 9.3%, respectively, and the D _PS was 13.9%, 10.8% and 9.7%, respectively. Clearly, the D _PD and D _PS of the flat pins both maximized at the initial pick-up angle of 18º, which were 3.2% and 2.9% smaller than the destruction index, so the root lumps were more easily destroyed. At the initial pick-up angle of 24º, the D _PD and D _PS were both over 7.1% smaller than the destruction index for both the circular pins and the flat pins, so the possibility of root lump destruction was lower. The destruction of root lumps can be eliminated by increasing the initial pick-up angle. However, when the initial pick-up angle surpassed the hole wall included angle and to acquire certain quan­tity of contraction, the insertion depth of pick-up pins should be shortened, which would reduce the clamping force and move the fracturable area upwards, and the destruction of root lumps would be more severe. One efficient solution is to change the direction of pin contraction, so that the pins would contract from the diagonal line of the root lump sections to the center, which would enlarge the initial pick-up angle and increase the distance and clamping force between the pins ( Han et al. , 2015 ; Jiang et al. , 2017 ).

At the same pick-up conditions, the D _PD and D _PS of flat pins always surpassed those of circular pins (Fig. 12). As for the reasons, the circular pins were in round face contact, and the raising substrate would slip bilaterally during the pin contraction, which would weaken the shearing and compression on the substrate caused by the pins. On the contrary, the flat pins were in plane contact with the substrate, which would more promote the migration of the substrate, and the new pore aggregation and crack formation in the root lumps would be more significant ( Karmakar et al. , 2007 ; Mojtaba et al. , 2013 ). The destruction of root lumps could be diminished by optimizing the sectional shape of pick-up pins, such as ellipse and semi-circle.

Generally, there is a group of optimal pick-up pa­rameters that could reduce the pore aggregation and crack expansion in the root lumps. The possibility of root lump destruction over the circular pins minimized when the pin diameter was 2 mm and the initial pick-up angle was 24º.

To validate the reasonableness of pick-up parameters as-selected, we used the automatic transplant structure to test seedling pick-up (Fig. 13A). When this structure worked, the feeding tray device and the gripper moved synchronously, which ensured the tomato plug seedlings were taken out of the trays one-by-one. The swing of the gripper was controlled by a circular cylinder, so the tomato plug seedlings were taken out of the trays and devoted into the cast seedling cups. The gripper was a 2-finger 4-pin structure (Fig. 13B), and the contraction and stretching of the pick-up pins were controlled by a squared cylinder. During the trials, the seedling pick-up rate was 45 plants/min.

Trials were conducted in accordance with Table 7. In each group of trials involving 512 plants, either pin diameter d or initial pick-up angle θ was altered. When the mass loss exceeded 25%, the root lump was considered as severely damaged ( Choi et al. , 2002 ).

The root lump destruction rate R was calculated as follows: