https://orcid.org/0000-0002-9395-6858
ABSTRACT
Grading and quality assessment is an important aspect of post-harvest management in pomegranate fruits. In India, the quality assessment is usually performed manually by feeling the fruit in hand. This manual assessment poses a lot of disadvantages such as uncertainty, tediousness, time consumption etc. Moreover there are no well-organized grading systems for testing quality of pomegranates. Aim of the present research work is to eliminate such problems associated with manual quality assessment by incorporating Machine Intelligence and Digital Image Processing techniques. The present work precisely redefines the new quality parameters associated with the existing grading criteria. The research work also proposes a unique Effective Quality Assessment (EQA) algorithm comprising of a holistic approach towards the grading and quality assessment of pomegranate fruits. Results of the research work are found to be 97.83% by using Artificial Neural Networks.
Introduction
India is one of the largest producers of pomegranate (Punica granatum) fruits. It is a highly remunerative crop to replace subsistence farming. It has the ability to withstand harsh and hostile climate there by making it a suitable crop for arid and semi-arid regions. It has immense medicinal values and yields higher remuneration with low cost of cultivation. The fruit is consumed fresh, that is for table purpose, or used as processed products viz., syrup, juice, jam, jelly carbonated beverage, wine etc.
The scope of export of Indian pomegranates is stretched to Qatar, Bangladesh, Bahrain, Saudi Arabia, Canada, Germany, United Kingdom, Japan, Kuwait, Sri Lanka, Omen, Pakistan, Singapore, Switzerland, U.A.E. and U.S.A. (APEDA, Citation2015). There has been a sturdy increase in area and production of pomegranate in the country. Export of the pomegranates is expected to increase by 6.97-fold and production by 10-fold by 2015 (Babu et al., Citation2012). In order to achieve these targets, sustained efforts are needed by all aspects concerned with pomegranate research and development including cultivation, post-harvest and processing techniques.
Grading for Post-Harvest Management in Pomegranates
Grading is an important operation of post-harvest management. It is done so as to obtain a reasonable price in the domestic or export market. Grading is usually performed on the basis of weight, size and external rind appearance. But in India, improper handling leads to a spoilage loss of 25–30% of the pomegranates there by reducing the profit margin of growers. Also, there are no well-organized marketing systems for testing the quality of pomegranates (Babu et al., Citation2012). Farmers send their fruit produce to the contractors who are then accountable for transporting it to distant markets (Benagi et al., Citation2009). Hence there is a vital need to propose a working system for quality assurance of pomegranate fruits post-harvest.
The field of Machine Vision has proved to be capable technology in minor and major industries. It discovers applications in various domains such as quality control, industrial process control, medical diagnostics, aerial surveillance, robotics, remote sensing, optical character recognition, face recognition, voice recognition, etc. With the improvements in the field of Digital Image Processing and Intelligent Control technologies, Machine Vision is extensively used in agriculture. Contrasting to industrial products, quality inspection of agricultural products offers certain challenges as the ‘appearance’ aspect is inconsistent and imprecise (Deepa and Geethalakshmi, Citation2011). Food industry is amidst the top 10 industries that widely make use of machine vision. Its role is inimitable in the field of automated sorting and grading of agricultural, horticultural and food products.
Major Contributions of the Work
With the motto of developing a non-destructive quality assessment system for post-harvest handling of pomegranates, the major research contributions of the present work are as follows:
making use of machine vision system in the process of grading and quality assessment apart from physical parameters;
redefining the existing grades for pomegranates and defining new quality values for each grade;
design of a holistic three-tier process in assessing the quality of the pomegranate fruits.
Organization of the Manuscript
The manuscript discusses literature review in section 2. Detailed materials and methods are presented in section 3. Section 3 also discusses the new quality definitions of the pomegranate grades. Section 4 discusses the results obtained in detail by providing the sample testing and comparison to previous research works. Finally the manuscript is concluded in section 5.
Acronyms Used in the Manuscript
Literature Survey
There are various researchers working around the clock on the aspect of post-harvest grading of various agricultural and horticultural produce. Lee et al. (Citation2008) presented a novel approach for quality evaluation of date fruits. Color quantization and color analysis techniques have been adopted in order to evaluate the quality of Medjool dates with an accuracy of 92.5%. Font et al. (Citation2014) investigated and verified the Nectarine varieties with an accuracy of 87%. Grading of the pal oil fresh fruit bunches was done automatically by Jamil et al. (Citation2009) by applying neuro fuzzy systems and obtained an accuracy of 73.3%. Almond images were segmented by employing artificial neural networks by Teimouri et al. (Citation2014) with an accuracy of 98.82%. Unay and Gosselin (Citation2005) developed a machine vision system to grade Jonagold apples and achieved a high accuracy of 90.3% with support vector machines. Narendra and Hareesha (Citation2011) developed a machine vision system to classify cashew kernels into six classes and obtained an overall accuracy of 80% with back propagation neural networks. Rocha et al. (Citation2010) introduced a machine system for multi-class fruits/vegetables classification in the super market scale using a unique feature and classifier fusion technique and obtained a classification error of 15%. Clement et al. (Citation2013) classified cucumbers according to the European Grading Standards and achieved a classification accuracy of 99%. Mustafa et al. (Citation2009) applied fuzzy logic in devising a new technique to perform automatic sorting and grading of fruits and observed the encouraging results.
Since the studies on pomegranate fruits grading are scarce, there is more scope in carrying out the research on grading and quality assessment of pomegranates. This is also in line with the industrial application of grading the pomegranate fruits. Hence the present research work is aimed at designing a novel algorithm to pop out the effective quality grade of pomegranates with the help of machine intelligence and digital image processing.
Materials and Methods
Fruit Sample Collection
The pomegranate fruits for the purpose of experimentation are collected a fresh from the farm fields. provides the details of the samples collected. depicts a sample image of each grade
Table 1. Details of pomegranate sample collection.
Figure 1. Samples images. (a) Grade 1, Weight-369gm, Diameter-92.37mm (b) Grade 2, Weight-280gm, Diameter-82.485868mm(c) Grade 3, Weight-198gm, Diameter-69.73944694mm.
Description of the Grades
provides descriptive characteristics of each of the grades of the pomegranate fruits. The acceptable tolerance in the table indicates the percentage of deviation from the standard characteristics (Benagi et al., Citation2009; Pomegranate, Citation2014).
Table 2. Description of the grades.
Overview of Grading and Quality Assessment
Grading is an important step for export and local markets that provides farmers the reasonable pricing. Fruits are usually graded on the basis of size, weight and external color features. The step by step procedure in grading and quality assessment has been depicted in terms of a block diagram in .
Figure 2. Overview of grading and qualification.
Following sections discuss each of the blocks shown in in detail.
Primary Level Grading Based on Fruit Size
The first parameter considered in grading pomegranate fruits is the size of the fruit. Grading is usually done with a human expert. After maturity on the plants, the fruits are harvested and a large bulk quantity of fruits is dumped in front of the human expert and his task is to pick each fruit and grade it and place it into the appropriate grading box. In this regard the first thing the human expert does is he/she picks up the fruits and depending on the feel of the fruit size in hand he/she will put the fruit into appropriate grading box. A common scenario of this kind is depicted in . Therefore if at all we would like to automate the process of grading, the first thing that we need to do is to use image size/diameter as the parameter to grade them.
Figure 3. Grading pomegranates with human expert.
Secondary Level Grading Based on Fruit Weight
The second level of grading to be performed is based on the weight as the parameter. This is because; weight acts as one of the important physical parameter in grading the pomegranates. The market price of the fruits is decided on the basis of weight. Hence it is essential to consider weight as the parameter.
Tertiary Level Grading Based on Image Analysis
Color and texture of the fruit plays a vital role in deciding the consumer grade of the fruit and hence the market price. Even if the size and weight are appropriate, sometimes it may so happen that the fruit may have cracks, spits, borers, not having attractive color and so on. Therefore, in the process of mimicking the fruit packing lines of industrial automation, the analysis of the external visual quality aspects is very much necessary. The methodology adopted in the present work is outlined in . The details regarding the steps to be followed can be found at our previous works at Kumar et al. (Citation2018). However a quick review of the same has been presented in the following section 3.2.3.1.
Figure 4. Steps in effective quality assessment.
A Brief Foray into Tertiary Level Grading
Images of pomegranates are acquired using a closed metal compartment. Steps in pre-processing include segmentation and histogram equalization followed by wavelet denoising. A total of 134 features are extracted consisting of 19 spatial domain features and 115 wavelet transform features. The description of the features extracted for each image is provided in . A feed forward back propagation neural network is used for the purpose of training the extracted features with 134 input neurons, 20 hidden neurons and 12 output neurons. Finally, the effective quality of a test sample can be displayed using the trained neural network.
Table 3. Description of the features extracted for each image.
The 12 output neurons correspond to the 12 classes of effective quality. The designations of the class labels are presented in
Table 4. Designations of the class labels.
The implementation is carried out using Matlab R2017a run on a Laptop with Intel Core2 Duo 2.2GHz processor and 3GB RAM
Effective Fruit Quality Assessment
Indeed, we are at an important junction of the research work that cannot be found in the previous literature of pomegranate fruits. The effectiveness of the present work can be augmented by beginning with a little explanation on what is known as the Effective Fruit Quality.
We have standard guidelines to be followed in order to grade the pomegranate fruits (Pomegranate, Citation2014). Up on following these guidelines and taking proper measures, we can successfully classify the fruits into different quality grades. But it is not always the case that fruits will be graded soon after harvesting. Many times farmers need to transport them to the nearest local market. Due to various conditions of the local environment, fruits may undergo gradation process after certain days of time. If we are not taking into account the number of days passed after harvesting, then we are wrongly/poorly interpreting the gradation results. If, for example, we have graded a fruit as belonging to G2, after two days of harvesting, the quality of the fruit however deteriorates fast as the number of days after harvest increases. Therefore as per the room temperature if fruit deteriorates fast, then the customer who purchases this fruit will not purchase the fruits from the same farmer next time.
Hence, a computer mediated decision support system is needed at this intersection in order to accurately predict the number of days passed after harvest so that the purchasing customer can make his decisions accordingly. That is, we need to present the customer the effective fruit quality rather than just the grade value. This has the following potential advantages:
We can effectively segregate the fruits.
We can provide the true market value for the fruits.
Leads to increased customer satisfaction.
Maintains the monetary stability of the farmer.
Describing the Effective Qualities
Assessment of the effective fruit quality of pomegranates is carried out by preserving the pomegranates for 8 days under room temperature and pressure. The fruits are analyzed with an interval of two days, thereby having four qualifications for each grade viz. Q1, Q2, Q3 and Q4, Q1 is being the best quality and Q4 being the lowest quality. The method of preservation for each of the quality categories is under room temperature (Avg. 25°C) and atmospheric pressure (1.01325 bar) with relative humidity (50%). provides the description and applications of each quality category in each grade. The quality definitions are same for each of the three grades i.e., for example the fruits belonging to G1Q1 or G2Q1 or G3Q1 all bear the same visual characteristics and applications of Q1 except physical characteristics.
Table 5. Description of the effective qualities.
Results and Analysis
demonstrates the results obtained after the analysis for 8 days for the grade G1. For the purpose of presentation of the results in the manuscript, only 19 features have been summarized in the table. Out of 19 features, 2 are physical parameters, 12 are color features and 5 are texture features. The results obtained for the grades G2 and G3 are summarized in Appendix. The method of preservation for each of the quality categories is under room temperature (Avg. 25°C) and atmospheric pressure (1.01325 bar) with relative humidity (50%). The table provides range of values of various parameters. All the results are in consistent with the fruit samples collected.
Table 6. Values of various parameters for qualities in Grade G1.
Interpretation of the Results Obtained
From it can be observed that if the diameter of a fruit falls in the range 85.39665779 to 105 and if weight is in the range 303–392 then it can be classified as belonging to Grade G1. However these are not the only two parameters that decide the quality of the fruit when we automate the process of grading and quality assessment. The visual parameters play a vital role in this regard. Consider for example a fruit belonging to grade G1. Its Rm value, if it lies in the range 136.6691 to 138.6525, then it can be classified as Q1. Similar interpretation holds good for the remaining features in order to decide the fruit as belonging to a particular quality. However in practical situations, it may so happen that a fruit may bear good weight and size but it may be rotten from inside or it might have been kept for longer days. In other words the fruit is of a lower quality. In such situations the quality of the fruit and hence the arils inside the fruit, is decided on the basis of the analysis of the visual parameters. This is because, if the fruit is rotten or kept for many days the color of the fruit changes toward dark red and texture becomes rough with dents. Consider the above same case where a fruit belongs to grade G1. Suppose if its value of Rm is in the range128.8521–132.6831 it may belong to quality Q3. Similar interpretation holds good for the remaining features. However, the fruit may also belong to the categories of G3Q2 or G3Q3 (as per the Rm values in and ). The value of the final result depends on all the 134 features as per the neural network training.
Hence, with the present research work it is possible to determine the exact quality within the specific grade, called as the effective quality, instead of just the grade of the fruit. Since there is a consideration of the aspects of size, weight and visual parameters, the proposed research work is a holistic approach toward the quality assessment of pomegranate fruits.
Results of Ann Training
The neural network training is performed for the number of epochs = 458. The trained neural network is used to predict the class of each of the input. Results of the training are shown as a confusion plot in . As it can be observed from the table, out of 150 G1Q2 fruits, 138 are correctly classified, 6 are misclassified as G1Q3 and 6 are misclassified as G1Q4. Out of 150 G1Q3 fruits, 144 are correctly classified, 6 are misclassified as G1Q4. Out of 150 G1Q4 fruits, 144 are correctly classified, 6 are misclassified as G2Q2. Out of 150 G2Q2 fruits, 145 are correctly classified, 5 are misclassified as G2Q3. Out of 150 G2Q3 fruits, 145 are correctly classified, 5 are misclassified as G2Q2. Out of 150 G3Q4 fruits, 145 are correctly classified, 5 are misclassified as G3Q1. Testing of the trained networks showed that out of 1800 samples 1761 samples are correctly classified yielding to an overall result of 97.83% accuracy.
Table 7. Confusion plot of results.
The Proposed EQA Algorithm
Whenever a fruit sample is brought for testing or as the fruits move on a conveyor belt in the industrial process lines, following algorithm can be applied to find their effective quality and segregate them.
EQA Algorithm
Step 1: Train the Neural Networks by using 134 feature values available in the database
Step 2: Primary Level Grading: Determine the fruit grade on the basis of diameter and volume by referring , and .
Step 2: Secondary Level Grading: Determine the fruit grade on the basis of weight by referring , and .
Step 3: Tertiary Level Grading:
Apply the steps as indicated in to obtain image features.
Feed the image features as input to the neural networks trained in Step 1.
Determine the effective quality and grade of the fruit sample.
Analysis of the Results with respect to Sample Testing of Pomegranate Fruits
shows the results of a sample fruit tested over a period of 8 days.
Table 8. Sample fruit testing results.
Interpretation of the Results
As observed from the above set of results, the transition from quality Q1 to Q4 can be seen by the gradual decrease in the values of Rm, Sm, Vm and Bm which stand for the red component, whiteness component, lightness component and blue-yellow transition component respectively. From this it can be inferred that the quality, which is a function of the above four color components, reduces as they reduce consecutively. Another important physical parameter which shows the reduction of quality over time is weight as seen in the table.
Another set of important observations that can be done from the above table is that the quality goes on reducing from Q1 to Q4 as the values of Bluem Cbm, which stand for blue component and blue difference component respectively, go on increasing. These components indicate the increase in levels of decay as time passes.
Random Sample Testing
Let us consider a random sample fruit. This fruit is not part of the training set. shows a random sample and its corresponding result. When this sample is collected randomly, its diameter is found to be 83.2173mm weighing 255gm. But the fruit is also having black patches with dark red color and rotten from inside. If at all we have automated the grading process as per the standards and guideless of APEDA and NRCP (), the fruit is found to be of grade G2 but it is not possible to determine its quality because the image features are not considered in the existing standards in assessing the quality. Therefore, if the image features are considered and fed to the trained neural networks as per the proposed EQA algorithm, then the result is G2Q4, that is, this fruit can neither be transported nor exported.
Table 9. Novelty of the present research work.
Figure 5. Random sample fruit for testing and its result.
In a nutshell, the previous standards could only determine the grade irrespective of the quality, in other words no matter whether the fruit is of best quality or worst, the fruit would be graded. This technique of grading could not help in deciding what application it suits. But in the proposed research work the grade is not irrespective of the quality which means that a lower grade higher quality fruit will not be misclassified as a better quality fruit in higher grade.
Novelty of Eqa Algorithm and Present Research Work
describes the novelty of EQA algorithm and hence that of present research work by comparing with the existing methods for grading pomegranates.
Comparison to Previous Works
The proposed research work is compared against few of the previous research works that concentrated on grading of various other fruits. Gist of the comparison is outlined in .
Table 10. Comparison to previous research works.
Conclusion and Future Scope
The present research work is aimed at quality grading of pomegranate fruits with the help of machine intelligence and digital image processing. Pomegranate fruits are collected from the farm fields. The fruits are stored under the room temperature and analyzed for eight days. Analysis is carried out with an interval of two days there by defining four qualities of each grade. The research work proposes new quality definitions for the existing grading criteria. A novel EQA algorithm is presented in order to assess the Effective Quality of the fruits. Digital image processing techniques are applied in order preprocess the captured images of pomegranates and a total of 134 features are extracted. Artificial neural networks are used to train the samples collected. Results of the testing indicated a performance of 97.83%. With the new quality definitions of each grade of the fruits and promising results, the present research work provides a holistic approach toward post-harvest handling of the pomegranates.
Future research directions include analyzing the fruit qualities with storage under STP and NTP conditions.
References
- APEDA. 2015. Pomegranate, p. 20–22. In: Study on identification of export oriented integrated infrastructure for agri products from Maharashtra & Gujarat. New Delhi, India: Agriculture Produce Export Development Authority.
- Babu, K.D., R.A. Marathe, and V.T. Jadhav. 2012. Post harvest management of pomegranate. ICAR – National Research Centre on Pomegranate, Solapur, India.
- Benagi, V.I., V. Nargund, R. Balikai, and M. Ravikumar. 2009. Pomegranate – Identification and management of diseases, insect pests and disorders. Dharwad, India: University of Agricultural Sciences
- Clement, J., N. Novas, J.A. Gazquez, and F. Manzano-Agugliaro. 2013. An active contour computer algorithm for the classification of cucumbers. Comput. Electron. Agric. 92:75–81. doi: 10.1016/j.compag.2013.01.006.
- Deepa, P., and S.N. Geethalakshmi. 2011. Improved watershed segmentation for apple fruit grading. Process Automation, Control and Computing (PACC), 2011 Int. Conf., Coimbatore, India: IEEE
- Font, D., M. Tresanchez, T. Pallejà, M. Teixidó, D. Mar-Tinez, J. Moreno, and J. Palacín. 2014. An image processing method for in-line nectarine variety verification based on the comparison of skin fea-ture histogram vectors. Comput. Electron. Agric. 102:112–119. doi: 10.1016/j.compag.2014.01.013.
- Jamil, N., A. Mohamed, and S. Abdullah. 2009. Automated grading of palm oil fresh fruit bunches (FFB) using neuro-fuzzy technique, p. 245–249. In: Proc. Int. Conf. Soft Comput. Pattern Recognit., IEEE. doi: 10.1109/socpar.2009.57.
- Kumar, R.A., V. Rajpurohit, and B. Jirage. 2018. Pomegranate fruit quality assessment using machine intelligence and wavelet features. J. Hortic. Res. 26(1):53–60. doi: 10.2478/johr-2018-0006.
- Lee, D.J., Y. Chang, J.K. Archibald, and C.G. Greco. 2008. Color quantization and image analysis for automated fruit quality evaluation, p. 194–199. In: Autom. Sci. Eng., 2008. CASE 2008. Washington, USA: IEEE Int. Conf., IEEE, Aug.
- Mustafa, N.B.A., S.K. Ahmed, Z. Ali, W.B. Yit, A.A.Z. Abidin, and Z.A.M. Sharrif 2009. Agricultural produce sorting and grading using support vector machines and fuzzy logic, p. 391–396. In: Proc. Int. Conf. Signal and Image Process. Appl., IEEE. doi: 10.1109/icsipa.2009.5478684.
- Narendra, V.G., and K.S. Hareesha. 2011. Cashew kernels classification using colour features. Int. J. Mach. Intelli. 3(2):52–57.
- POMEGRANATE: Cultivation, Marketing and Utilization. 2014. Technical Bulletin No. NRCP/2014/1, Solapur, India: ICAR-National research center on pomegranate.
- Rocha, A., D.C. Hauagge, J. Wainer, and S. Goldenstein. 2010. Automatic fruit and vegetable classification from images. Comput. Electron. Agric. 70(1):96–104. doi: 10.1016/j.compag.2009.09.002.
- Teimouri, N., M. Omid, K. Mollazade, and A. Rajabipour. 2014. A novel artificial neural networks assisted segmentation algorithm for discriminating almond nut and shell from background and shadow. Comput. Electron. Agric. 105:34–43. doi: 10.1016/j.compag.2014.04.008.
- Unay, D., and B. Gosselin (2005). Artificial neural network-based segmentation and apple grading by machine vision, p. II–630. In: Image Process., 2005. ICIP 2005. Genoa, Italy: IEEE Int. Conf.. Vol. 2. IEEE, Sep.
Appendix
Table A1. Values of various parameters for qualities in Grade G2.
Table A2. Values of various parameters for qualities in Grade G3.