**5. Proxy-detection image processing for seed development and germination performance of chicory achenes by chlorophyll fluorescence**

Several aspects of seed quality are tested routinely to minimize the risk of sowing seedlots that do not have the capacity to produce the desired crop. Amongst these tests, seed germination is important since it represents the percentage of pure seeds that have the potential to produce established seedlings in the field. The rate of germination corresponds to the reciprocal of the time needed for a given germination percentage to be reached (Halmer, 2008). Accurate procedure of germination tests performed in laboratory is defined by the International Seed Testing Association (ISTA).

Texture, Color and Frequential Proxy-Detection

Destain, 2011).

seeds is about 6 %).

**5.2 Results** 

Image Processing for Crop Characterization in a Context of Precision Agriculture 65

Fig. 17. The components of the chlorophyll imaging device and the main steps of image analysis. (1) Xenon lamp with blue optical bandpass filter (370 - 450 nm), (2) optical fiber, (3) cone of blue light, (4) seeds from one capitulum, (5) blue light reflected by the seeds, (6) chlorophyll fluorescence (650 - 730 nm), (7) highpass optical filter (665 nm), (8) CCD camera with zoom objective, (9) connection to the computer, (10) raw image, (11) background correction, then creation of individual images, (12) automated detection of the pappus using dedicated software and estimation of the levels of fluorescence FPER and FPAP (Ooms &

The observed period corresponds to the phase of reserve deposition in the seed described in Bewley & Black (1994). Figure 18 shows the evolution of CF during this phase on the stalk, the weight parameters (dry weight DW and water content WC) and the germination performance (GP and GR). The dry weight increased, while the water content was still high at the end of the observed period (> 45 % on the stalk at 44 DAF, while the WC of stored

Besides these laboratory tests, seed processing lines includes cleaning machines that remove dust and waste material, and conditioning machines performing dimensional sizing, density sorting, and colour sorting. In conventional colour sorting, discoloured seeds are rejected, on basis of inspecting seeds individually to detect differences in reflected colour.

However, even after several sorting operations, some seed batches can contain a large proportion of viable seeds but still not sufficient for commercial use. These batches are lost because the viable and non-viable seeds cannot be separated using the conventional processing methods. The proportion of immature seeds in these lost batches is unknown. It would therefore be useful to provide a new, non-destructive method of distinguishing immature seeds from mature seeds in order to improve sorting processes.

In this context, the potential of fluorescence imaging (Chen et al., 2002; Nedbal & Whitmarsh, 2004) has been examined. The chlorophyll degrades during fruit ripening and the process of degradation was described by Barry (2009). The chlorophyll is also a highly fluorescent molecule. Fluorescence occurs when some of the light absorbed by the chlorophylls is re-emitted at longer wavelength, typically between 650 and 750 nm. The fluorescent properties of chlorophylls have been used to evaluate the maturity of cabbage seeds (*Brassica oleracea* L.) (Jalink et al., 1998; Jalink et al., 1999). The results showed that the magnitude of the chlorophyll fluorescence (CF) signal was inversely related to the quality of seeds. The relationship between the CF and germination performance was studied for tomato (*Solanum lycopersicum* L.) by Jalink et al. (1999) as in their previous study (Jalink et al., 1998). They concluded that seeds with an intermediate CF level were of the best quality, followed by seeds having a low CF signal. Seeds having a high CF signal were the worst. Konstantinova et al. (2002) measured the CF of barley grains (*Hordeum vulgare* L.) with a SeedScan I Laser Sorter (Satake, Stafford, TX. USA), using the principle developed by Jalink et al. (1998) but including a laser light source instead of a LED. They concluded that sorting a barley seed lot into six subsamples varying in CF values resulted in an optimal quality for the subsamples with low and intermediate CF signals. Suhartanto (2002) thoroughly described the relationships between the fruit CF, seeds CF and germination performance of tomato.

#### **5.1 Image analysis**

A specific image analysis code was developed with the GNU Octave language (Ooms & Destain, 2011). After applying background correction (the fluorescence values were divided by the reflectance signal of paper and multiplied by 60), the images were segmented and images of individual seeds were created, each of them being rotated along the main axis of the seed. The pappus side, which is brighter and larger than the radicle tip, was automatically detected on the basis of the mean width of the left half, its mean fluorescence intensity, the right half width and the right half fluorescence intensity. The accuracy of the detection was greater than 98%. The image was thereafter divided into the "pericarp zone" (Pe, 77% of the seed length) and the "pappus zone" (Pa, 23% of the length). The value of 77% was a compromise based on the observation of 100 seed images. The mean fluorescence values of the two zones were recorded for data analysis. The measurement system and image analysis are summarised in figure 17.

Fig. 17. The components of the chlorophyll imaging device and the main steps of image analysis. (1) Xenon lamp with blue optical bandpass filter (370 - 450 nm), (2) optical fiber, (3) cone of blue light, (4) seeds from one capitulum, (5) blue light reflected by the seeds, (6) chlorophyll fluorescence (650 - 730 nm), (7) highpass optical filter (665 nm), (8) CCD camera with zoom objective, (9) connection to the computer, (10) raw image, (11) background correction, then creation of individual images, (12) automated detection of the pappus using dedicated software and estimation of the levels of fluorescence FPER and FPAP (Ooms & Destain, 2011).

## **5.2 Results**

64 Agricultural Science

Besides these laboratory tests, seed processing lines includes cleaning machines that remove dust and waste material, and conditioning machines performing dimensional sizing, density sorting, and colour sorting. In conventional colour sorting, discoloured seeds are rejected, on basis of inspecting seeds individually to detect differences in

However, even after several sorting operations, some seed batches can contain a large proportion of viable seeds but still not sufficient for commercial use. These batches are lost because the viable and non-viable seeds cannot be separated using the conventional processing methods. The proportion of immature seeds in these lost batches is unknown. It would therefore be useful to provide a new, non-destructive method of distinguishing

In this context, the potential of fluorescence imaging (Chen et al., 2002; Nedbal & Whitmarsh, 2004) has been examined. The chlorophyll degrades during fruit ripening and the process of degradation was described by Barry (2009). The chlorophyll is also a highly fluorescent molecule. Fluorescence occurs when some of the light absorbed by the chlorophylls is re-emitted at longer wavelength, typically between 650 and 750 nm. The fluorescent properties of chlorophylls have been used to evaluate the maturity of cabbage seeds (*Brassica oleracea* L.) (Jalink et al., 1998; Jalink et al., 1999). The results showed that the magnitude of the chlorophyll fluorescence (CF) signal was inversely related to the quality of seeds. The relationship between the CF and germination performance was studied for tomato (*Solanum lycopersicum* L.) by Jalink et al. (1999) as in their previous study (Jalink et al., 1998). They concluded that seeds with an intermediate CF level were of the best quality, followed by seeds having a low CF signal. Seeds having a high CF signal were the worst. Konstantinova et al. (2002) measured the CF of barley grains (*Hordeum vulgare* L.) with a SeedScan I Laser Sorter (Satake, Stafford, TX. USA), using the principle developed by Jalink et al. (1998) but including a laser light source instead of a LED. They concluded that sorting a barley seed lot into six subsamples varying in CF values resulted in an optimal quality for the subsamples with low and intermediate CF signals. Suhartanto (2002) thoroughly described the relationships between the fruit CF, seeds CF and germination performance of

A specific image analysis code was developed with the GNU Octave language (Ooms & Destain, 2011). After applying background correction (the fluorescence values were divided by the reflectance signal of paper and multiplied by 60), the images were segmented and images of individual seeds were created, each of them being rotated along the main axis of the seed. The pappus side, which is brighter and larger than the radicle tip, was automatically detected on the basis of the mean width of the left half, its mean fluorescence intensity, the right half width and the right half fluorescence intensity. The accuracy of the detection was greater than 98%. The image was thereafter divided into the "pericarp zone" (Pe, 77% of the seed length) and the "pappus zone" (Pa, 23% of the length). The value of 77% was a compromise based on the observation of 100 seed images. The mean fluorescence values of the two zones were recorded for data analysis. The measurement system and

immature seeds from mature seeds in order to improve sorting processes.

reflected colour.

tomato.

**5.1 Image analysis** 

image analysis are summarised in figure 17.

The observed period corresponds to the phase of reserve deposition in the seed described in Bewley & Black (1994). Figure 18 shows the evolution of CF during this phase on the stalk, the weight parameters (dry weight DW and water content WC) and the germination performance (GP and GR). The dry weight increased, while the water content was still high at the end of the observed period (> 45 % on the stalk at 44 DAF, while the WC of stored seeds is about 6 %).

Texture, Color and Frequential Proxy-Detection

better artefacts elimination.

implementation.

**7. References** 

Image Processing for Crop Characterization in a Context of Precision Agriculture 67

yield, the number of wheat ears has to be determined and is the result of a detection and a counting step. Detection has been done by developing specific image acquisition system and by implementing algorithms like colour-texture hybrid space or more classical image processing based on Fourier filtering, according to the frequential information (redundant information) included in the images. In this chapter, we presented the high pass Fourier filtering technique which gives satisfying and robust wheat ear detection with lower computing time. Moreover, we have compared the detection results with those obtained by representation in a hybrid space. Even if satisfying results are obtained for this qualitative experiment, some improvements should be done such as including an automatic threshold determination after the inverse FFT and an efficient shape analysis in order to obtain a finer wheat ear detection and

This work has also revealed the possibility of extending the use of pattern recognition techniques and textural feature analysis to other applications, including the automatic determination of the wheat growth stage, aiming at creating a decision tool for farmers. At this stage, it seems that we should include an agronomic validation in order to propose more specific model to help farmers. In the same context one perspective could be to couple proxy-detection with satellite or aerial images. Moreover using multispectral data would

The study of hyperspectral imaging demonstrates the potential to separate weed-culture. A more comprehensive study must now be conducted to assess the robustness and the spectral and spatial minimum necessary resolutions in the context of an operational

Most fruits and seeds have measurable levels of chlorophyll, respectively in their pericarp or testa, and this chlorophyll degrades with time. In the case of chicory, the commercial seed is a fruit which cannot be hulled and the seed is not observable directly. The low amount of 10 chlorophyll implies the use of a highly sensitive device and because of the presence of a distinct pappus, the imaging of fluorescence is favourable. It would be profitable to identify the factors (variety, season, climate, hydric stress, etc.) influencing the evolution of CF to predict the characteristics of the decrease of CF and to predict if its decrease is always

Finally, combination of several image acquisition systems should give more interesting

Aparicio, N., Villegas, D., Casadesus, J., Araus, J.L., & Royo, C. (2000). Spectral vegetation

Barry, C.S. (2009). The stay-green revolution: Recent progress in deciphering the

Bewley J., & Black M. (1994). Seeds: physiology of development and germination. *Plenum*,

indices as non-destructive tools for determining durum wheat yield. *Agronomy* 

mechanisms of chlorophyll next term previous term degradation next term in

results and open the door of the detection of other important crop characteristics.

probably improve even more the efficiency of such an approach.

concomitant with the increase of the germination performance.

higher plants. *Plant Science*, 176(3), 325-333.

*Journal*, Vol. 92, pp. 83–91.

ISBN 0-306-44748-7.

Fig. 18. Evolution of chlorophyll fluorescence (FPER: pericarp, FPAP: pappus), dry weight, water content at harvest, germination percentage and germination rate of chicory seeds at each maturation duration on the stalk from 16 to 44 days after flowering. Means with confidence intervals of the means.

The two following facts are in favour of the use of CF features for the differentiation of immature chicory seeds from mature ones, and as indicators of seed vigour:


On the other hand, the efficiency of CF features may be negatively affected by the large variability of individual measurements (random differences between individuals).

Current work aims at estimating the correlations between the CF features, the weight parameters and the germination variables in outdoors and greenhouse environments, and to assess the added value of CF features in comparison to weight, size and density features to distinguish between viable and non-viable seeds using sorting simulations.

## **6. Conclusion**

To predict wheat yield or to determine wheat growth stages, remote sensing is not the only solution. Proxy-detection systems allow to acquire high resolution images to be treated by robust algorithms such as high pass Fourier filtering. For example, to predict yield, the number of wheat ears has to be determined and is the result of a detection and a counting step. Detection has been done by developing specific image acquisition system and by implementing algorithms like colour-texture hybrid space or more classical image processing based on Fourier filtering, according to the frequential information (redundant information) included in the images. In this chapter, we presented the high pass Fourier filtering technique which gives satisfying and robust wheat ear detection with lower computing time. Moreover, we have compared the detection results with those obtained by representation in a hybrid space. Even if satisfying results are obtained for this qualitative experiment, some improvements should be done such as including an automatic threshold determination after the inverse FFT and an efficient shape analysis in order to obtain a finer wheat ear detection and better artefacts elimination.

This work has also revealed the possibility of extending the use of pattern recognition techniques and textural feature analysis to other applications, including the automatic determination of the wheat growth stage, aiming at creating a decision tool for farmers. At this stage, it seems that we should include an agronomic validation in order to propose more specific model to help farmers. In the same context one perspective could be to couple proxy-detection with satellite or aerial images. Moreover using multispectral data would probably improve even more the efficiency of such an approach.

The study of hyperspectral imaging demonstrates the potential to separate weed-culture. A more comprehensive study must now be conducted to assess the robustness and the spectral and spatial minimum necessary resolutions in the context of an operational implementation.

Most fruits and seeds have measurable levels of chlorophyll, respectively in their pericarp or testa, and this chlorophyll degrades with time. In the case of chicory, the commercial seed is a fruit which cannot be hulled and the seed is not observable directly. The low amount of 10 chlorophyll implies the use of a highly sensitive device and because of the presence of a distinct pappus, the imaging of fluorescence is favourable. It would be profitable to identify the factors (variety, season, climate, hydric stress, etc.) influencing the evolution of CF to predict the characteristics of the decrease of CF and to predict if its decrease is always concomitant with the increase of the germination performance.

Finally, combination of several image acquisition systems should give more interesting results and open the door of the detection of other important crop characteristics.
