Importance of Image Processing Technology in Agriculture

1. Introduction

Advances in Technology has shown effectivity in the implementation of automated solutions in many fields. As the technology evolves its use grows exponentially and can be seem all over the globe in various applications.

Studies reveal that by 2050 humanity will need 70 per cent more food than we have today (Vasudevan, Kumar, & Bhuvaneswari, 2016). Something must be done as food is wasted not only because the human manual labor still the most used resource but also because the environment has been paying a high price for the humanity mistakes.

Vasudevan et al. (2016) proposed a solution that they named as “Precision Farm” where the use of Unmanned Aerial Vehicles and Unmanned Ground Vehicles are used to collect data to make an efficient use of water, fertilizers and pesticides.

(Rokhmana, 2015) presents practical applications of Unmanned Aerial Vehicle (UAV) in Indonesia that succeeded in obtaining information from the environment via image processing to help farmers with tasks such as vegetation monitoring, assessment of plant healthy, land preparation and stock evaluation.  Their system could collect images with space resolution <10cm resulting on Orthophoto images and Digital Elevation Models(DEM).

The dissemination of solutions like these is important not only from the business perspective but also from humanity survival perspective as with the growth of our species the resources available will be more and more scarce and we will need smarter solutions to avoid humanity from extinction (Vasudevan et al., 2016).

Studies have revealed that at least 70 per cent of the Indian economy relies on Agriculture and that it is important for the country economy to find a solution for the diseases that affects the crops decreasing the waste and increasing the incomes for agricultural producers (Tripathi & Maktedar, 2016).  Khirade and Patil (2015) explains the India population is dependent of agriculture and that diseases on plants reduces the quality and quantity of products affecting the economy of the country.

The vast biodiversity encountered in our planet is estimated to be about 10 million species of plants but the classification and identification of this enormous biodiversity is time consuming and error-prone but at the same time necessary to our evolution (Mata-Montero & Carranza-Rojas, 2016).

Technology is being heavily used in Agriculture for the past years as the case study demonstrated by Cao et al. (2000) where a robot was used to harvest strawberries in a farm. Their system used a DOF’s Cartesian coordinate manipulator to locate and extract features of the fruit using two color CCD cameras. The proposed solution processed images and converted it from RGB to L*a*b* extracting features that could be used to find the fruit position, orientation and shape.

In one of their papers, Vasudevan et al. (2016) stated:   “It is concluded from the above that technological applications are the solution to an increased agricultural production required to feed the growing population with the limited available resources.”.

(Gonzalez-de-Soto, Emmi, Benavides, Garcia, & Gonzalez-de-Santos, 2016) achieved a reduction of 50% on emissions of pollutant gases on the atmosphere modifying a tractor to use hybrid energy system which consists on a hydrogen cell, small pumps and small fans.

The humanity is living in a century where the increase of CPU power and the development of new machine learning algorithms is a crucial factor for the adoption of these technologies for the agricultural field as it makes the implementation of a solution cheaper making the return of investment valuable. Marfatia (2013) explains that machine learning can be effective if combine with existing recognition methods and describe the training process applied on machines to develop a neural network capable of extracting valuable information from farm products images.

A combination of machine learning, robots and IOT devices like the paper presented by  (Mohanraj, Ashokumar, & Naren, 2016) that explore series of modules can reduce wastage and increase profit on farms businesses.

2. Different Technologies

This section presents solutions that have been proposed by researchers and that demonstrated satisfactory level of acceptance by the farmers.

2.1. Internet of Things

The Internet of Things known as IOT was defined in Recommendation ITU-T Y.2060 (06/2012) and it forms a network of physical devices that can be controlled via software creating a virtual network between objects such as: devices, vehicles and sensors (Kapoor, Bhat, Shidnal, & Mehra, 2016).

A research made by Wilf et al. (2016) shows quite impressive average levels of recognition reaching of 70% which demonstrated how effective machine learning can be recognizing leaves via its shapes other properties such as texture and color.

Kapoor et al. (2016) shows that a combination of cloud computing and IOT devices is more effective than previous solutions and that companies that started using IOT devices back on 2014 are already seeing the benefits of its use. They have been applied these devices for asset tracking, security, field force management, energy data management and condition based monitoring and it has been widely used in the agriculture field.

2.2. Unmanned Aerial Vehicles

(Rokhmana, 2015) explains that Indonesia’s technological agricultural market has been increasing along the years with the inclusion of Unmanned Aerial Vehicles (UAV) as a replacement to satellite imagery as the resolution of a UAV is superior to the ones generated by satellites.  They present a low-cost solution that uses a two thousand dollars UAV that is easy to operate by farm’s staff. The application of their solution in a Palm-Oil plantation that succeeded calculating stand per Hectare (SPH) and individual palm tree via Orthophotos. Although their solution had the capability to count the trees they agreed that improvements on their system such as accuracy need to reach levels of 95% to be acceptable by their clients. They reached a level of accuracy of 15 cm and their solution was well accepted at a sugar factory field to calculate the taxation for their sugar cane asset.

UAV have been explored and showing effectivity collecting data from the vegetation of farms. It shows that a low-cost UAV carrying a digital pocket camera can collect data used to count products and help farms to calculate their annual expenses.

2.3. Unmanned Ground Vehicles

To reduce emissions of greenhouse gases such as CO₂ and NO_x emitted by tractors and other pollutants that can cause healthy problem such as worsen respiratory diseases Gonzalez-de-Soto et al. (2016) proposed a modification on the commercial CNHi Boomer 3050 CVT compact tractor power system which reduces the emission of these gases significantly. They also included a control system which enables the vehicle to control itself using an unmanned system making it become one great Unmanned Ground Vehicle (UGV) solution. Equipped with an ICE, PV panel, batteries and a hydrogen fuel cell(HFC) it proved to be an effective use of technologies based on clean energy source reducing pollutant emissions by 50%.

2.4. Machine Vision

The last 10 years have been marked by the advances of new solutions for pineapple quality grading and only 20-10% of the pineapple harvest made in Thailand is freshly consumed whereas 80-90% is processed in factories after passing through a manual quality grading system (Suksawat & Komkum, 2015).

Looking for the clients demands, Mustafa et al. (2008) succeed on their experiment where size and quality of bananas and percentage of ripeness were measured by software.

The importance of creating mechanisms to improve the quality of the classification of these fruits is important not only from an economical perspective but also from a standardized process. (Suksawat & Komkum, 2015) explains that the quality of the fruits is classified by external and internal features and that human factors influence on the manual process which makes it unreliable.

2.4.1. Disease recognition

Due to the natural human low visual capabilities it is not possible to detect the first symptoms of plant diseases as the occur at a microscopic level so that the necessity of using technology as an auxiliary to improve the accuracy and efficiently recognize diseases on plants at the early stages of it (Pujari, Yakkundimath, & Byadgi, 2015).

The amount of work necessary to identify plant diseases manually is the main reason it is important to create a process where this could be identified automatically via an artificial intelligence (Khirade & Patil, 2015).

The intention of this research is to further explore the different Image Processing methods and its applications on the Agriculture sector exploring current proposals that will lead humanity to a more technological approach aiming to a safer environment and demonstrate the importance of adopting technological based systems in our society.

3. Case Studies

It is widely known that the use of any kind of technological based system is avoided not only because of the cost that it may cause or the expensive learning curve of it but also because there is a lack of awareness of what would be an appropriate use for it in each specific case. As technology advances, the developing of new strategies that improve farm outputs such as Irrigations, Fertilizers, pesticides and quality yield ceased to be an issue for adoption after the dissemination of Image Processing as a tool that is used as an auxiliary to reduce costs and provide an acceptable level of accuracy while recognizing pests and plants (Vibhute, K. Bodhe, Vibhute, & K. Bodhe, 2012).

3.1. Internet of Things

(Mohanraj et al., 2016) explains that farms have the necessity of the inclusion of technology in their lands to keep their business profitable and that India is one of the first countries to implement advanced solutions in this sector. To make India’s business more profitable they proposed a solution using a TI CC3200 Launchpad interconnected with multiple sensors that can measure many aspects of the land being studied which also reduced labor cost on their lands.

The solution tested by (Mohanraj et al., 2016) used a system developed with Microsoft .Net Framework which facilitates the development of solutions that need to communicate via internet such as webpages, in this case they used ASP.NET with C# and XML as their main providers of data. Their system could communicate via smart sensors which would transfer GPS coordinates to the central computer. The monitoring of the field was done via RFID and sensors what communicated with CGM modules. All these pieces of software working on a three-layered architecture hosted in the cloud giving access to the farmers via Wi-fi or 3G.

Mohanraj et al. (2016) explains that technologies such as GPRS, GPS, 3G and Wi-fi are not reliable and are a challenge when delivering real time solution to the farmers and to suppress this deficiency they created a Bluetooth interaction allowing the system to work in offline mode.

The system proposed by (Mohanraj et al., 2016) has modules that are send relevant information to the farmers in real time such as:

Reminder module that reminds reaping, fertilizer application, pesticides spraying and irrigation timings via SMS notification.

Monitoring Plant Growth which measures the plant’s height and evaluates the plants development.

Irrigation Planner which checks the soil moisture and based on information collected from IOT devices decide the crop water necessity maintaining a continuous flow in the field irrigation system.

Crop Profit Calculator which evaluates the selling price of the crop based on its features.

Calamities check which uses cloud technologies to access relevant data from the Yahoo weather webservices and monitor unusual activities such as firing in the field.

Problem identifier which checks whether the irrigation system is working as expected or if any of the peripherals have problem such as power supply.

Calculation of Water need which works in parallel with IOT devices to check whether the soil needs more water combined with webservices from the Yahoo webservices to check the local weather conditions.

Optic Monitor which measure the sunlight using a Light Intensity Sensor which guarantees that the plant will grow accordingly.

Well Dry Check and Field Dry Check which evaluates the soil moisture to decide where it is dry or not.

In average Farmers have poor knowledge of Technology, so new solution cannot rely on their knowledge to keep working. Also, it is not affordable to have experts to keep the farming working as expected. Modules like the ones presented by Mohanraj et al. (2016) shown its efficacy and reliability as it delivers real-time information to the clients.

3.2. Image Processing

In India 70% of the country rely on Agriculture as the main economy’s motor, it is essential for the constant growth of the country that the disease detection and diagnostics processes occur in a fast pace (Choudhar & Gulati, 2015). Image processing as a facilitator would increase the level of adoption of technological based systems as it can contribute in many aspects of the Agriculture sector. The use of machine vision has proven to be effective on detecting disease on fruits, insects and food grading. Weeds can be harmful from a farm perspective as it competes with crop for water, light, nutrients and space, so the development of different methods based in Image processing is very popular and it uses techniques such as detection of edges, color and other attributes of weeds to recognize it (Vibhute et al., 2012).

The classification method used by Khirade and Patil (2015) follow five steps: Image Acquisition, Image pre-processing, Image Segmentation, Feature Extraction and Detection and Classification.

After acquiring the images using a camera, the image is preprocessed using algorithms such as: Image clipping, Image smoothing and Image enhancement which removes noises of it before the application of a grayscale filter used to create a histogram using a distribution function to enhance the image.

The segmentation occurs in three level:

• Boundary and spot detection where RGB image to create a HIS model where the boundaries and spots of infections are detected with the help of boundary detection algorithm and eight connectivity of pixels.
• K-means clustering where a set of features of the leaf is scanned seeking for the minimum number of squares between the object and the cluster being detected.
• Otsu Threshold where binary images are generated using greyscale and setting pixels below or above some threshold level to 0 and 1 respectively.

The next step is the feature extraction where features such as color, texture and edges are extracted from the images.

After the extraction is concluded the process continues for two paths: Neural Networks or Back Propagation.

3.2.1. Fruit Grading System

(Mustafa et al., 2008) explains that banana is a very important product to the Malaysia’s economy as it is the second most popular product there. To meet the client’s necessity, they proposed a solution where bananas were classified by its size: extra-large, large, medium and small and the percentage of its ripeness by its color: green and yellow.

The most popular detection approach is the edge detection but even passed more than 20 years after its creation it still have its particularities (Mustafa et al., 2008) . The edge detection approach can be easily implemented via MATLAB’s built in functions for image processing(Mustafa et al., 2008).  Their experiment uses a coin as a reference object to measure the size of each fruit.

To detect the ripeness percentage of the fruit they classified it into six stages: unripe, pre-mature, almost mature, almost ripe, ripe and too ripe and checked the presence and intensity of yellow on the fruit.

Higher fruit and food standards can be achieved using a grading system says Vibhute et al. (2012)  and concludes saying that the environment can be saved if the use of herbicides is controlled via an Imaging Processing Application as the recognition of weeds in specific areas will show exactly were to apply them correctly.

3.3. Precision Farm

Vasudevan et al. (2016) explain that by 2050 humanity will need 70% more food than we have today, so to address this problem they proposed a solution that they named as “Precision Farm” where the use of Unmanned Aerial Vehicles and Unmanned Ground Vehicles are used to collect data to make an efficient use of water, fertilizers and pesticides. In one of their papers they stated:   “It is concluded from the above that technological applications are the solution to an increased agricultural production required to feed the growing population with the limited available resources.”. It can be seen then, that the dissemination of these solutions is important not only from the business perspective but also from the survival perspective of it. As the population expand the resources available will be more and more scarce and we will need smarter solutions to avoid humanity from extinction.

3.4. Machine Learning

Ranked the first country in the production of agricultural products such as milk, crops, cotton, vegetables and fruits, India relies on agriculture to guarantee its survival.

The use of machine vision in agriculture has shown to be effective as it can be seen on many applications around India. Crops are managed via ground and aerial sensors that send images to devices that analysis and process it to identify diseases in leaf, stem, fruits and vegetables. Vegetables and fruits are classified to increase the quality of the products.

Marfatia (2013) shows that technology can be present and effective in many parts of the agricultural field such as: Pest management for the detection of insects, crop assessment for identification of unwanted weeds, quality and grading for classification of fruits and vegetables, harvesting of fruits and vegetables, estimation of plant nitrogen and real-time object tracking.

Marfatia (2013)  presents a few classifiers that are important for having a machine learning implemented such as Linear Discriminant Classifier (LDC) which tries to locate a boundary between the data of the image and is used to extract features of the products.

The Nearest neighbor classifier (k-NN) uses fruit features such as color, shape and are used to train the system using a Euclidean algorithm that looks for a K shortest distance to the input fruit and assign it to the closes K from its origin.

Support Vector Machines (SVM) can used to classify both linear and nonlinear data. It was used to identify weeds and determine the areas that needed to be sprayed with pesticides.

Artificial Neural Network (ANN) is one of the most popular methods and uses many well-known approaches as its inspiration and works as a biological nervous system which makes it robust while dealing with substantial amounts of data. ANN has shown impressive results for the classification of fruits like apples that were classified into categories (AA, A, B, C, D) representing its quality standards.

Rule Based System, also known as fuzzy system defines a set of rules that are used to extract data from the images, extract features of the images, the construction of this rules and the recognition and classification of the fruits.

The results presented by Marfatia (2013) experiments shows that training neural networks has resulted in a high level of accuracy reaching 75%-96% whereas feature extraction techniques such as fractal analysis and CIELAB had impressive 100% accuracy levels. In average common methods such as color mapping, histogram had reached 85%-97%.

Machine learning has been demonstrated its value and efficacy combining existing methods and providing expressive results for fruit grading and sorting.

4. Applications

Wilf et al. (2016) explains that one of the most challenging problems in botany is that there is such a variety of complex shapes on plants that any level attempt to recognize them via computer would be worth it. They succeeded with their experiment with an elevated level of accuracy having a system that could recognize and classify images into families and orders using a computer vision system over a database of 7597 leaf images from 2001 genera which works on heat maps that shows the location of various informative leaf characters known as novels.

Vasudevan et al. (2016) has shown that the use of Unmanned Aerial and Ground Vehicles is effective for farm management using image processing algorithms that are applied on the pictures taken from it. To collect these high definition images from the crop, field a framework for robot development called D. Robot Operating System was used to control the devices which were responsible on taking pictures of the field and creating a reflectance map of crops using RGB (Red, Green and Blue), NIR (near infrared), RE (rededge), multi SPEC 4C (multispectral), thermo MAP (thermal infrared) sensor/cameras they could collect and generate relevant information to the farmers.

Tripathi and Maktedar (2016) explain that robots were used to capture images from the farmlands and processed via a disease detection system that works on five steps: input, processing, segmentation, extraction of features, disease classification and the identification of the disease. Machine vision techniques were used to facilitate the analysis and images found on the internet helped to train the machine to increase the recognition rates.

On an attempt of finding the best approach for plants disease detection Tripathi and Maktedar (2016) developed many solutions that could be used for disease detection with a very high level of accuracy as can be seen on their study for detection on wheat crops which has shown accuracy of 95% for classification of diseases after a test over 269 images of vegetables.

Computer Vision based solutions used by Tripathi and Maktedar (2016) are effective on the classification and detection of diseases on fruits, vegetables and other products.

Tripathi and Maktedar (2016) states that a Support Vector Machine (SVM) which is the best approach of all the solutions they tested showing accuracy of 93.79% on tests over 120 images taken via mobile cameras.

Mata-Montero and Carranza-Rojas (2016) explain that previous methods of identification have shown its effectivity such as: single-access key, interactive keys, morphometric approaches, crowd sourcing and DNA barcoding but the use of machine learning has had enormous progress as technologies such as OCR, biometrics and optical sorting has changed the way botanists classify these organisms.

Kapoor et al. (2016) conducted an experiment that uses MATHLAB software in combination with IOT devices to collect data such as temperature, humidity, soil moisture and light density to understand how it affects the health of the plant.

Kapoor et al. (2016) built a solution using a Soil Moisture sensor, a Temperature and Humidity sensor, a serial JPEG camera and a SD card shield assembled on an Arduino interface which basically collects data from the soil and take a picture and saves it in the SD card.

The experiment conducted by Kapoor et al. (2016) was able to assess the health of a plant using image processing techniques combined with IOT devices which collected data that would fed their system and find out the necessities of the plants being studied.

Kapoor et al. (2016) explains that the IOT sensors can be easily installed on Unmanned Ground Vehicles or Unmanned Aerial Vehicles which makes the monitoring of the land possible as these devices can send data through CDMA/GSM protocol.  Thus, a reduction of errors while applying pesticides or mineral to the plants and land as the Unmanned Vehicles can carry on the right amount of these substances and put it on the soil or applying I on the plant with more precision saving money and giving to the soil and plants a special treatment.

The solution presented by Suksawat and Komkum (2015) uses a picture of the pineapple which is transformed into a gray scale version of it to facilitate the recognition of the edges of the fruit which allows the calculation of its size and weight. The tests were run through three types of pineapple: Nanglae, Sriacha and Phuket which presented elevated levels of accuracy of identification of the weight and size reaching 87.5% of success.  The margin of error was very low reaching 2.30% and 5.24% for size and weight respectively.

Another method used by Suksawat and Komkum (2015) was an artificial neural network which resulted on a level of accuracy of 83% whilst the computer vision method was able to classify the fruits with 100% of accuracy. The latest method uses a mixture of color analysis that enhance the image and compares its color with a database giving the highest level of accuracy.

Pujari et al. (2015) developed a methodology capable of identify and classify fungal diseases on plants. They used algorithms that extracted information from the pictures taken from the plants and classified the plants based on this data. The experiment was well succeeded with fruits, vegetables and cereals crops which had symptoms of the fungal disease. Their identification methods focused on statistic methods that were capable of quantitatively detect and classify a fungal disease using Gray Level Co-occurrence Matrix (GLCM), Gray Level Run Length Matrix (GLRLM) and Nearest Neighbor (NN) techniques which resulted on levels of accuracy of 91.37% for GLCM and 86.715% for GLRLM. After applying their block wise algorithm there was an increase to 94.085% of accuracy for the classification of these diseases.

5. Conclusions

This paper focused on the importance of the adoption of image processing technologies for the soil quality, detection of plant diseases and for the quality grading of fruits as the agricultural field has faced many challenges over the years trying to solve this issue and the previous solutions are becoming obsoletes with the increase of technology approaches that are more effective.

The literature for this area is extensive and lead us to experiments where machine learning approaches the use of heat maps of plants images to identity and classify leaves. It was shown, that it is possible to increase the quantity and the quality of products harvested on the farm with the help of image processing techniques and that this work done manually is not a viable solution it would take too long to be done and would be much more expensive.

Internet of Things (IOT) devices are being extensively used to collect data that helps identifying the correct use of pesticides and fertilizers on the farmlands and classifying these products to follow quality standards defined by governments via the use of quality grading of the fruits.

Terrestrial and Aerial Unmanned Vehicles are being used to capture high definition images of plants and soil samples to provide data to a framework created for robots that is used to construct a real time virtual environment using Simultaneous Localization and Mapping algorithms which is a crucial factor for the detection of the plants necessity maintaining them health and increasing the quality of the products harvested on the farm.

When seem from outside the solutions presented here look like science fiction, but the reality is that it is happening everywhere right now and the numbers has shown us that the adopters of these technologies are being rewarded for their investment.

6. References

Cao, Q., Nagata, M., Gejima, Y., Shrestha, B. P., Hiyoshi, K., & Ootsu, K. (2000). Basic Study on Strawberry Harvesting Robot (Part I). IFAC Proceedings Volumes, 33(29), 49-54. doi:http://dx.doi.org/10.1016/S1474-6670(17)36750-2

Choudhar, G. M., & Gulati, V. (2015). Advances in Image Processing for Detection of Plant Diseases.

Gonzalez-de-Soto, M., Emmi, L., Benavides, C., Garcia, I., & Gonzalez-de-Santos, P. (2016). Reducing air pollution with hybrid-powered robotic tractors for precision agriculture. Biosystems Engineering, 143, 79-94. doi:https://doi.org/10.1016/j.biosystemseng.2016.01.008

Kapoor, A., Bhat, S. I., Shidnal, S., & Mehra, A. (2016, 6-8 Oct. 2016). Implementation of IoT (Internet of Things) and Image processing in smart agriculture. Paper presented at the 2016 International Conference on Computation System and Information Technology for Sustainable Solutions (CSITSS).

Khirade, S. D., & Patil, A. B. (2015, 26-27 Feb. 2015). Plant Disease Detection Using Image Processing. Paper presented at the 2015 International Conference on Computing Communication Control and Automation.

Marfatia, R. P. a. S. N. a. R. (2013). Image Processing and Machine Learning for Automated Fruit Grading System: A Technical Review. International Journal of Computer Applications, 81, 29-39.

Mata-Montero, E., & Carranza-Rojas, J. (2016). Automated Plant Species Identification: Challenges and Opportunities. In F. J. Mata & A. Pont (Eds.), ICT for Promoting Human Development and Protecting the Environment: 6th IFIP World Information Technology Forum, WITFOR 2016, San José, Costa Rica, September 12-14, 2016, Proceedings (pp. 26-36). Cham: Springer International Publishing.

Mohanraj, I., Ashokumar, K., & Naren, J. (2016). Field Monitoring and Automation Using IOT in Agriculture Domain. Procedia Computer Science, 93, 931-939. doi:http://dx.doi.org/10.1016/j.procs.2016.07.275

Mustafa, N. B. A., Fuad, N. A., Ahmed, S. K., Abidin, A. A. Z., Ali, Z., Yit, W. B., & Sharrif, Z. A. M. (2008, 26-28 Aug. 2008). Image processing of an agriculture produce: Determination of size and ripeness of a banana. Paper presented at the 2008 International Symposium on Information Technology.

Pujari, J. D., Yakkundimath, R., & Byadgi, A. S. (2015). Image Processing Based Detection of Fungal Diseases in Plants. Procedia Computer Science, 46, 1802-1808. doi:http://dx.doi.org/10.1016/j.procs.2015.02.137

Rokhmana, C. A. (2015). The Potential of UAV-based Remote Sensing for Supporting Precision Agriculture in Indonesia. Procedia Environmental Sciences, 24, 245-253. doi:http://dx.doi.org/10.1016/j.proenv.2015.03.032

Suksawat, B., & Komkum, P. (2015, 20-22 May 2015). Pineapple quality grading using image processing and fuzzy logic based on Thai Agriculture Standards. Paper presented at the 2015 International Conference on Control, Automation and Robotics.

Tripathi, M. K., & Maktedar, D. D. (2016, 12-13 Aug. 2016). Recent machine learning based approaches for disease detection and classification of agricultural products. Paper presented at the 2016 International Conference on Computing Communication Control and automation (ICCUBEA).

Vasudevan, A., Kumar, D. A., & Bhuvaneswari, N. S. (2016, 15-16 July 2016). Precision farming using unmanned aerial and ground vehicles. Paper presented at the 2016 IEEE Technological Innovations in ICT for Agriculture and Rural Development (TIAR).

Vibhute, A., K. Bodhe, S., Vibhute, A., & K. Bodhe, S. (2012). Applications of Image Processing in Agriculture: A Survey. International Journal of Computer Applications, 52(2), 34-40. doi:10.5120/8176-1495

Wilf, P., Zhang, S., Chikkerur, S., Little, S. A., Wing, S. L., & Serre, T. (2016). Computer vision cracks the leaf code. Proceedings of the National Academy of Sciences, 113(12), 3305-3310. doi:10.1073/pnas.1524473113