Recognition and detection of pig posture based on instance segmentation and computer vision in pig farm

레자나심 ( Nasim Reza ),카비르사자둘 ( Sazzadul Kabir ),하케아스라쿨 ( Asrakul Haque ),정선옥 ( Sun-ok Chung ) 한국농업기계학회 2022 한국농업기계학회 학술발표논문집 Vol.27 No.1

Pig posture changes throughout the growing period are most often indicators to illness. Monitoring pigs postural movements enables us to identify morphological changes in pigs early and to detect potential risk factors for pig health. Large-scale pig farming requires extensive manual monitoring by pig farmers, which is time-consuming and laborious. Computer vision-based monitoring of posture activities over time may help to limit the spread of disease infections. The objective of this study was to recognize and detect pig posture using an masked based instance segmentation in the pig farm. Two automatic video acquisition systems were installed from top and side view, respectively. RGB images were extracted from the RGB video files and used for annotation work. Manual annotation of 200 images were prepared as training dataset, including the three postures: standing, lying, and eating from bin. An instance segmentation framework was employed to recognize and detect pig posture. A region proposal network is used in the first stage of the masked R-CNN based instance segmentation procedure. It obtains features from candidate boxes using RoIPool and conducts classification and bounding-box regression in the second step. Proposed method was evaluated using test image dataset and the experimental results showed the proposed framework obtained a F1 score of 0.911. Our work investigated a new way for recognizing and detecting pig posture in pig farm, which enables useful research into vision-based, real-time automated pig monitoring and diseases assessment.

Basic performance test for sound detection and remote monitoring in Pig Farm

레자나심 ( Nasim Reza ),초두리밀론 ( Milon Chowdhury ),키라가샤피크 ( Shafik Kiraga ),정선옥 ( Sun-ok Chung ) 한국농업기계학회 2021 한국농업기계학회 학술발표논문집 Vol.26 No.2

Precision livestock farming is an intelligent technology, which allows the closer monitoring of each animal on farms. Sound based precision farming provides considerable benefits compared to other technology, such as imaging sensor, motion sensors, etc. In addition, sound sensors are inexpensive, no direct contact, and a huge number of animal can be observed using a single sensor. The objective of this study was to investigate a remotely monitored sound detection and imaging system in pig farm for early detection of respiratory diseases. Three microphones and three RGB cameras with three micro-controller were used to receive the sound and image data in the pig farm. Total 30 pigs were covered by our surveillance system. A sound analysis algorithm was developed to record the sounds received by the microphones and distinguished the pig sounds from the outside noises. The sound was then processed by the algorithm to detect the abnormal sound of pigs. The images were synchronized and used to monitor the unwanted movement and behaviour. High, medium, and low frequency sounds were detected. The results showed that the detection efficiency for high frequency sound was around 85%, and for low frequency sound was 73%. Moreover, movement of pigs were also monitored by images. From this study, it would be feasible to recognize early respiratory illness in pigs through automated and sequential monitoring of sounds and images within the pig farm.

Real-time sound monitoring using 2D convolutional neural network (CNN) for pig diseases symptoms detection in pig farm

레자나심 ( Nasim Reza ),하케아스라쿨 ( Asrakul Haque ),카림레자울 ( Rejaul Karim ),송민호 ( Minho Song ),김국환 ( Gookhwan Kim ),정선옥 ( Sun-ok Chung ) 한국농업기계학회 2022 한국농업기계학회 학술발표논문집 Vol.27 No.2

Monitoring and preventing diseases in livestock is essential for modern farming, and an early warning system may significantly reduce the economic impact of diseased events. Manual monitoring of pigs in a pig farm is time consuming and labor intensive. Automatic monitoring for pig diseases may help to control the spread of infections. The purpose of this research was to identify the signs of sickness in pigs using acoustic monitoring in real-time. Two microphones were installed in the pig farm for automatic sound acquisition. The sound signals were converted into spectrograms by fast fourier transform (FFT) and mel-frequency cepstral coefficients (MFCC) as a characteristic parameter. Using a 2D convolutional neural network (CNN) and features extracted from the spectrogram, we presented a classification approach for real-time application. The conversion of sound inputs into spectrograms made it possible to recognize by the use of CNN. For the real-time detection, the proposed algorithm showed the ability to recognize the sounds of cough, sneezing, scream, and crushing with an overall recognition accuracy of 75.8%, 69.5%, 72.4%, and 71.6%, respectively, and an average F1-score of 81.2%. Future work is needed to enhance sound detection robustness.

Image based algorithm for growth prediction of pennywort plant grown in a plant factory

이슬람수마이야 ( Sumaiya Islam ),레자나심 ( Nasim Reza ),초두리밀론 ( Milon Chowdhury ),키라가샤피크 ( Shafik Kiraga ),정선옥 ( Sun-ok Chung ) 한국농업기계학회 2021 한국농업기계학회 학술발표논문집 Vol.26 No.2

Plant growth prediction typically relies on the estimation of changes in plant structure and size. The leaf is one of the visual structures of plants, which has a significant impact on growth. The objective of this study was to predict pennywort plant growth using an image processing algorithm. Pennywort plant was grown in the plant factory, where ambient environmental variables were maintained precisely. The experiment was carried out for four weeks. RGB images of the plant were captured by a digital camera from the top of the plants everyday. In the image processing algorithm, the images were converted to grayscale and then binary masking was applied to classify each pixel as belonging to the region of interest. The masked images were segmented from the background. Then the region filling technique was applied to fill out the leaf region. We calculated the total pixel number in the image leaf area and calculated the leaf area using reference object. Actual plant leaf area was also continuously measured by a leaf area meter with specific time intervals without hindering plant growth. Our proposed algorithm demonstrated a high correlation of 0.954 between the actual and image-based leaf area measurements. A linear regression curve was found and growth was predicted using the desired cultivation period on the regression equation. Growth prediction model showed the potentiality to estimate plant growth cultivated in controlled environment.

Performance evaluation of a low-powered electric platform prototype under various agricultural purposes

알리모하마드 ( Mohammod Ali ),레자나심 ( Nasim Reza ),초두리밀론 ( Milon Chowdhury ),구란다즈아스라푸자만 ( Ashrafuzzaman Gulandaz ),키라가샤픽 ( Shafik Kiraga ),정선옥 ( Sun-ok Chung ) 한국농업기계학회 2021 한국농업기계학회 학술발표논문집 Vol.26 No.2

Multi-purpose platform of a machine performs various agricultural operations that can reduce labor and enhance the convenience of aged people and women. To obtain the maximum working efficiency and reliability of a prototype machine, a field experiment is required. Therefore, the purpose of the study was to evaluate the performance of a electric-drive tracked platform based on the power and pulling force analysis to conduct the major agricultural operations in different load and speed conditions. Grass mowing, and chemical spraying operations were selected as primary agricultural activities to evaluate the performance of the prototype tracked-platform. A data acquisition system was established with a torque sensor and a load cell to measure the maximum torque and pulling force. To record the working speeds of the platform, a global positioning system (GPS) was used. The field experiment was conducted on sandy clay loam soil condition on a 50 m operational path and the maximum speed was recorded up to 8 km/h (1 to 8 km/h). The average power requirements for a whole system (lawnmower + sprayer-trailer), a sprayer-trailer filled with 150 L of water, a lawnmower (without sprayer-trailer), and a platform itself were 1.27±0.35, 1.70±.13, 0.93±0.21, and 1.45±.19 kW, respectively. The maximum pulling force was measured 1.98±1.12, 0.614±0.46, and 19.28±11.32 kgf for hauling the rear trailer at 150L-payload condition, and 0.81±0.60, 0.416±.34, and 4.50±3.88 kgf at the unloaded condition in X, Y, and Z direction, respectively. The highest amount of average power and pulling force were recorded at maximum load condition. The power and pulling force requirements were fluctuated due to the effects of the different driving speeds. This study would help to provide information to the manufacturer for advance modifications of the tracked-platform.

Impact Of Variability In Dynamic And Solar Intensity On Ground Canopy Sensors Detecting Vegetation Indices

하케아스라쿨 ( Asrakul Haque ),레자나심 ( Nasim Reza ),하비네자엘리에젤 ( Eliezel Habineza ),강영호 ( Yeong Ho Kang ),이경도 ( Keong Do Lee ),정선옥 ( Sun-ok Chung ) 한국농업기계학회 2022 한국농업기계학회 학술발표논문집 Vol.27 No.2

The popularity of ground based sensors (active crop canopy sensors) is going upward for determining crop growth status and recommending additional fertilizer. By employing these sensors, Vegetation indices (VI) amplified the significance of identification in determining crop nutritional status. However, The sensor output might be continuously affected by operating and ambient factors such as movement speed and solar radiation respectively. In this study, we investigated the effects of movement speed and PAR (Photosynthetically Active Radiation) on the consistency of Crop Circle ACS-435 & DAS44X during NDVI (normalized differential vegetation index) measurement. The effects of these two parameters were evaluated on different platforms. Several movement speeds (0, 0.1, 0.2, 0.3, 0.4, and 0.5 ms-1) and PAR at different times of a sunny day with no cloud interference (10.00- 11.00, 13.00- 14.00, and 16.00- 17.00 hrs) were considered as factors to be assessed for enhancing the sensors' accuracy in plant health detection. In addition, linear regression models, Root mean square error (RMSE), and additional graphical analysis was employed to assess the effects of sensor movement speed and solar source intensity. A significant difference in speed was found in Crop-circle and DAS44X during the calculation of the NDVI. However, the difference in illumination intensity didn’t appear to have much effect on Crop-circle and DAS44X. Data acquisition has been shown to be most effective at speeds up to 0.3 ms-1 and at any time of day for both sensors in order to produce the highest output feasible. These findings illustrated the speed range certainty and daytime flexibility of Crop-Circle and DAS44X.

Control and arrangement of small-sized suspension-type dehumidifier for ICT based greenhouse environment

구란다즈아스라푸자만 ( Ashrafuzzaman Gulandaz ),레자나심 ( Nasim Reza ),초두리밀론 ( Milon Chowdhury ),키라가샤픽 ( Shafik Kiraga ),정선옥 ( Sun-ok Chung ) 한국농업기계학회 2021 한국농업기계학회 학술발표논문집 Vol.26 No.2

Humidity control inside greenhouses is essential for optimum plant growth and physiological disorders and diseases managements. The humidity response and variability depend extensively on the performance of the dehumidifier. The objective of this research was to evaluate the performance of a small-sized suspension-type dehumidifier in terms of temperature and humidity changes and spatial and vertical variability in a greenhouse. The dehumidifier consisted of a 0.6 kW compressor and a 0.1 kW fan. We compared the performance of dehumidifier among the different installation layouts like one at the center, two at the center (facing opposite directions from the center to the sides), one at either of the sides, two at both of the sides (facing to the center). To evaluate the functional ability of the dehumidifier, 45 temperature and humidity sensors were placed at three layers (top, middle, and bottom) and in five sections for monitoring the environmental status inside the greenhouse. Two additional sensors were placed in front of the dehumidifier and outside of the greenhouse. A wireless sensor network was used to collect the data for 90% to 70% dehumidifier operating conditions and monitor the humidity status during the operation and collected the data. The humidity response results showed that the time required for 90% to 70% dehumidification were 33 minutes. Temperature fluctuates 1degree after 11 minutes in upper layer and 27 minutes in bottom layer. The spatial and variability results indicated that the changes in humidity at, two at the center (facing opposite directions from the center to the sides) were higher than those in the other setup of dehumidifier in greenhouse. The outcomes of this research will be helpful for the development of low-power, small-sized dehumidifying systems and its setup position in greenhouse for cultivation.

농업용 붐 분무기의 노즐 높이 및 분무 균일 성 측정을 위한 초음파 센서에 대한 온도의 영향

샤피크키라가 ( Shafik Kiraga ),레자나심 ( Md Nasim Reza ),초두리밀런 ( Milon Chowdhury ),이슬람나피울 ( Md Nafiul Islam ),정선옥 ( Sun-ok Chung ) 한국농업기계학회 2021 한국농업기계학회 학술발표논문집 Vol.26 No.1

The increasing popularity of boom height controllers requires a desired nozzle height above the spray targets for effective spraying. The objective of this study was to determine the effect of temperature on nozzle height measurement using ultrasonic sensors and the impact on spray uniformity. Experiments were carried out with a constant nozzle spacing of 50 cm and a pressure of 600 kPa considering temperature compensation and non-compensation states of ultrasonic sensor measurements. Nozzle heights of 30, 50, and 70 cm were selected for reference. Two measurements were carried out for each selected height with a temperature compensated and non-compensated state at 15, 25, and 300C. The spray distribution was also determined for each measurements. Coefficient of Variation (CV) and percent measurement error (ME) were used to characterize spray uniformity and temperature effect on height measurement. The estimated heights with temperature compensation were closer to the reference heights, which exhibits a low ME. The ME increased from low to high temperatures. Temperature compensated heights resulted in more even spraying trends, with lower CV values accompanied by non-compensated heights. Our findings showed that, ultrasonic sensors need temperature correction for proper Nozzle height measurement and optimal spray distribution.

Working speed analysis of a gear-driven rotary planting mechanism of a 12-kW self-propelled riding-type automatic onion transplanter

초두리밀론 ( Milon Chowdhury ),레자나심 ( Nasim Reza ),알리모하마드 ( Mohammod Ali ),고란닺앗라퐂자만 ( Ashrafuzzaman Gulandaz ),권행주 ( Haing-ju Kwon ),정선옥 ( Sun-ok Chung ) 한국농업기계학회 2021 한국농업기계학회 학술발표논문집 Vol.26 No.2

The development of a riding-type automatic onion transplanter could be effective in improving the mechanization rate in onion cultivation, where the working speed analysis plays a vital role and determines planting performance and efficiency. The objective of this study was to select a suitable working speed of a 12 kW self-propelled riding-type automatic onion transplanter for efficient planting of onion seedlings with minimum damage. The proposed transplanting mechanism consisted of six assembling units of picking, conveyors, and rotary planting mechanism, where every rotary planting unit needs a continuous supply of onion seedlings at a certain rate for uniform and upright plantation. To enable the smooth collection and plantation of onion seedlings, analysis was carried out via a mathematical working trajectory model of the planting mechanism, virtual prototype simulation, and validation tests using a physical test bench. In the mathematical model analysis and simulation, the suitable rotational speed was found as 60 rpm and it was able to transplant 60 and 120 seedlings/min using the single and double unit assembly of planting mechanism, respectively. A 130 mm/s forward speed of the transplanter was preferable in terms of seedling uprightness and low damage. A forward speed of 130 mm/s with a transplanting speed of 120 seedlings/min was preferable in terms of achieving a high degree of seedling uprightness. A field test using the real prototype of the onion transplanter would be necessary to verify the accuracy of these findings.

Comparison of the performance of Ultrasonic sensors and single load cell impact plate for estimating the mass of Chinese Cabbage

샤피크키라가 ( Shafik Kiraga ),레자나심 ( Nasim Reza ),초두리밀런 ( Milon Chowdhury ),구란다즈아스라푸자만 ( Gulandaz Md Ashrafuzzaman ),정선옥 ( Sun-ok Chung ) 한국농업기계학회 2021 한국농업기계학회 학술발표논문집 Vol.26 No.2

Chinese cabbage is a commercially valuable crop due to its various uses. Among its important quality parameters is mass, which can be used in the development of yield maps for yield monitoring. Previous research focused mostly on the use of load cell(s) and stereo-based approaches for mass estimation. This study aimed to propose a new method of mass estimation using three HY-SRF05 ultrasonic sensors, and compare it’s performance to that of a single load cell under laboratory conditions. An impact plate was fabricated and installed to receive impact of Cabbage as it dropped off an inclined conveyor. The load cell was calibrated with different loads and achieved an R2 fit of 0.986. Cabbage mass was calculated from the load cell signals. The effects of different dropping heights, plate angles, and conveyor speeds were also investigated. On the same conveyor, three sensors, two installed opposite to each other and at the top, were first calibrated using known distances, and then used to measure Cabbage length and thickness. Cabbage mass was calculated from its volume assuming an elliptical shape.. The proposed method had a mass estimation accuracy greater than 91%, slightly lower than that for the load cell’s 95%. The proposed method showed potential for mass estimation.