Search

This paper introduces an autonomous self-propelled Napa cabbage harvester, designed to significantly improve the efficiency and effectiveness of the traditionally labor-intensive harvesting process. The harvester integrates three key modules: a cutting, an attitude control, and a loading module. The cutting module is equipped with an attitude control module that ensures precise severance of the Napa cabbage stems, minimizing damage to the crop and maintaining product quality. The attitude control module employs a backstepping-based force control that continuously adjusts the cutting angle and height to ensure consistent cutting precision, even on uneven terrain, thereby optimizing the quality of the Napa cabbages. The loading module automates the collection and transfer of harvested Napa cabbages into storage, significantly reducing the physical burden on workers and improving operational efficiency. Field experiments demonstrated improvements, including a 42–66% reduction in task time compared to manual harvesting, as well as a 37% increase in cutting accuracy through the use of autonomous control. The proposed system presents a comprehensive solution for enhancing productivity, reducing labor demands, and maintaining high crop quality in Napa cabbage harvesting, offering a practical approach to modernizing agricultural practices.

The mechanization of agricultural work has contributed significantly to the improvement of agricultural productivity and reduced production costs [...]

In this study, a regression model of paddy soil properties using diffuse reflectance spectroscopy was developed to replace chemical soil analysis as a more efficient alternative. Soil samples were collected and analyzed from saltwater paddy fields located in Jeongnam-myeon, Hwaseong-si, Gyeonggi-do in the Republic of Korea, and the spectral data of wet and dry soil were collected. The regression models were compared and analyzed using partial least squares regression (PLSR) with Savitzky–Golay smoothing (SG smoothing) and Standard Normal Variate (SNV) preprocessing to predict the soil properties. Analysis showed that the predictive regression model of wet soil with SG smoothing and an SNV did not meet the evaluation criteria of a fair model. However, the regression model of dry soil with SG smoothing was fair for clay, pH, EC, and TN at RPD = 1.90, 1.87, 1.60, and 1.43 and R2 = 0.79, 0.81, 0.64, and 0.64, respectively, while the regression model of dry soil with an SNV was good for clay, pH, EC, and TN at RPD = 2.21, 1.96, 1.70, and 1.44 and R2 = 0.84, 0.81, 0.76, 0.69, respectively. When developing predictive regression models of soil properties, the accuracy for dry soil was higher than that for wet soil, and when applying a single round of preprocessing, the regression model with SNV preprocessing was more accurate than that with SG smoothing.

The purpose of this study was to construct a map of expected fertilization rates for nitrogen (N) and phosphorus (P2O5) based on measurements of components in soil samples and to identify the spatial variabilities of four lots of a salt-affected paddy field in Korea. Four salt-affected paddy field lots in Korea were divided into 30 sectors for collecting soil samples. They were then analyzed for soil organic matter (SOM), silicon dioxide (SiO2), total nitrogen (TN), and available phosphorus (Av.P2O5) in accordance with international standards. Expected fertilization rates of N and P2O5 were developed as prescription standards for the application of fertilizer to paddy fields. They were derived using a model of the fertilization rates of N and P2O5. To determine the presence of spatial correlation and continuity in the given fields, a spherical variogram was used. Based on the spherical model with the application of a regular kriging interpolation, maps of the contents of TN and Av.P2O5 as well as the expected fertilization rates of N and P2O5 at each sector of 1×1 m2 were developed. The expected fertilization rate of N at each sector appeared in the range of min. 10.0 g to max. 25.7 g, while that of P2O5 appeared in the range of min. 0.68 g to max. 8.46 g.

The winnowing machine of chili pepper harvesters was developed to reduce the potential problem of low pepper stem and fruit separation. The developed winnowing machine was combined with two impellers and a center bearing to prevent a strain on the drive shaft and to ensure durability. The terminal velocity of chili pepper was measured, and an aerodynamic analysis was conducted based on this winnowing machine. A CFD (Computational Fluid Dynamics, Ansys Fluent 2020 R1) analysis was conducted for three levels of discharge port guide form (0, 3, and 5 guides) and three levels of rotating speed (1600, 1800, and 2000 RPM) of a winnowing machine designed utilizing aerodynamic analysis results. A validation test was conducted by fabricating a winnower test device. As for aerodynamic analysis conducted using measured values of terminal velocity, chili pepper fruits were collected at an outlet wind speed lower than 17.5 m/s and chili pepper branches were separated at a speed higher than 12.5 m/s. As a result of CFD analysis, the wind speed deviation at outlets of the 0-, 3-, and 5-guide depending on the rotating speed appeared to be 15.8, 1.4, and 1.0 m/s on average, respectively. The result of the CFD analysis showed values higher than wind speeds of the actual winnower test device by a minimum of 0 and a maximum of 2.4 m/s. Through the CFD analysis and the wind speed validation test of the winnower test device, optimal conditions to separate foreign materials were found to be a winnowing machine at a rotating speed of 1800 RPM with a discharge port having three guides or a winnowing machine at a rotating speed of 2000 RPM with a discharge port having five guides.

The chili pepper harvester has shown potential problems of low pepper stem separation and a high pepper damage rate. The low pepper stem separation has required additional labor, which consists of separating the pepper and stem after pepper harvesting. To improve the stem separation and sorting function of pepper harvesters, three-shaft and four-shaft drum-type stem-separation devices were manufactured, and performance tests were conducted to assess these devices. In an attempt to reduce the damage rate, a brush was used as the teeth in the drum-type stem-separation device. In the factor test, the rotational speeds of shaft 1(A), shaft 2(B), shaft 3(C), and the conveyor for the three-shaft drum were 0.9, 2.7, 1.3, and 0.5 m/s, respectively. The rotational speed of the four-shaft drum was the same as that of the three-shaft drum except for shaft 4(D), and the rotational speed of this additional D was set to 1.3 m/s, which was the same as that of C. In the non-moving status during the non-picking operation of the pepper harvester, the average stem-separation efficiency (SSE) of the four-shaft drum increased by 1.2%, the average pepper with twig rate (PTR) decreased by 5.9%, and the average damage rate (DR) increased by 3.7% compared to the three-shaft drum. In the moving status during the picking operation of the pepper harvester, the SSE of the four-shaft drum increased by 3.6%, the PTR decreased by 9.1%, and the DR increased by 3.8% compared to the three-shaft drum, so an improvement in the pepper stem-separation capacity was observed.

For harvest automation of sweet pepper, image recognition algorithms for differentiating each part of a sweet pepper plant were developed and performances of these algorithms were compared. An imaging system consisting of two cameras and six halogen lamps was built for sweet pepper image acquisition. For image analysis using the normalized difference vegetation index (NDVI), a band-pass filter in the range of 435 to 950 nm with a broad spectrum from visible light to infrared was used. K-means clustering and morphological skeletonization were used to classify sweet pepper parts to which the NDVI was applied. Scale-invariant feature transform (SIFT) and speeded-up robust features (SURFs) were used to figure out local features. Classification performances of a support vector machine (SVM) using the radial basis function kernel and backpropagation (BP) algorithm were compared to classify local SURFs of fruits, nodes, leaves, and suckers. Accuracies of the BP algorithm and the SVM for classifying local features were 95.96 and 63.75%, respectively. When the BP algorithm was used for classification of plant parts, the recognition success rate was 94.44% for fruits, 84.73% for nodes, 69.97% for leaves, and 84.34% for suckers. When CNN was used for classifying plant parts, the recognition success rate was 99.50% for fruits, 87.75% for nodes, 90.50% for leaves, and 87.25% for suckers.

Two prototypes of small chili pepper harvesters that attach to walking cultivators were designed and manufactured before field performance tests were conducted. The two prototypes were designed and manufactured with different main frame materials, forms of divider, picking guides, and helix rotation speeds. The maximum helix speed of the first prototype was 500 rpm, and the helix rotation speeds of the second prototype were a minimum of 510 rpm and a maximum of 730 rpm. Field performance tests were conducted on two species of chili, the AR Legend and the Jeokyoung, to determine which was suitable for mechanization. The Jeokyoung species was found to be most suitable for mechanization as its harvest efficiency was higher and its pepper left on plant rate and ground fall loss rate were lower than AR Legend’s. When the first and second prototypes were compared at helix rotation speeds of 500 to 510 rpm, in the case of the AR Legend, the average harvest efficiency of the second prototype was higher than the first prototype by 2.2%, the average pepper left on plant rate was lower by 2.1%, and the average ground fall loss rate was lower by 3.9%. In the case of the Jeokyoung, the performance of the second prototype was further improved over the first prototype as the average harvest efficiency increased to 5.2%, and the difference in average ground fall loss rate increased to 8.8%.

Accurate seismic ground response analysis is crucial for the design and safety of civil infrastructure and establishing effective mitigation measures against seismic risks and hazards. This is a complex process due to the nonlinear soil properties and complicated underground geometries. As a simplified approach, the one-dimensional wave propagation model, which assumes that seismic waves travel vertically through a horizontally layered medium, is widely adopted for its reasonable performance in many practical applications. This study explores the potential of sequence deep learning models, specifically 1D convolutional neural networks (1D-CNNs), long short-term memory (LSTM) networks, and transformers, as an alternative for seismic ground response modeling. Utilizing ground motion data from the Kiban Kyoshin Network (KiK-net), we train these models to predict ground surface acceleration response spectra based on bedrock motions. The performance of the data-driven models is compared with the conventional equivalent-linear analysis model, SHAKE2000. The results demonstrate that the deep learning models outperform the physics-based model across various sites, with the transformer model exhibiting the smallest average prediction error due to its ability to capture long-range dependencies. The 1D-CNN model also shows a promising performance, albeit with occasional higher errors than the other models. All the data-driven models exhibit efficient computation times of less than 0.4 s for estimation. These findings highlight the potential of sequence deep learning approaches for seismic ground response modeling.