Soil parameters are crucial aspects in increasing agricultural production. Even though Bangladesh is heavily dependent on agriculture, little research has been done regarding its automation. And a vital aspect of agricultural automation is predicting soil parameters. Generally, sensors relating to soil parameters are quite expensive and are often done in a controlled environment such as a greenhouse. However, a large scale implementation of such expensive sensors is not very feasible. This work tries to find an inexpensive solution towards predicting soil parameters such as soil moisture and temperature, both of which are crucial to the growth of crops. We focus on finding a robust relation between the above mentioned soil parameters with the nearby weather parameters such as humidity and temperature, irrespective of the weather. We apply different machine learning models like multilayer perceptron (MLP), random forest, etc. to predict the soil parameters, given the humidity and temperature of the surrounding environment. For all the experiments we have used a custom made dataset, which contains around 9000 datapoints of soil moisture & temperature, ambient humidity & temperature. The data has been collected in an uncontrolled agriculture bed via inexpensive sensors. Our results show that XGBoost regressor achieves the best results with an R2 score of 0.93 and 0.99 for soil moisture and soil temperature data respectively. This suggests very high correlation between the weather parameters and soil parameters. The model also portrayed a very low root mean squared error and mean absolute error of 0.037 & 0.015 for soil moisture and 0.001 & 0.0008 for soil temperature. Our results show that it is indeed possible to find the soil parameters from the corresponding weather, which will have great impact on mass agricultural automation. The dataset has been made publicly available at https://github.com/Nadimulhaque0403/Soil_parameter_prediction_dataset.

Nowadays, Convolution Neural Network (CNN) based deep learning methods are widely used in detecting and classifying fruits from faults, color and size characteristics. In this study, two different neural network model estimators are employed to detect apples using the Single-Shot Multibox Detection (SSD) Mobilenet and Faster Region-CNN (Faster R-CNN) model architectures, with the custom dataset generated from the red apple species. Each neural network model is trained with created dataset using 4000 apple images. With the trained model, apples are detected and counted autonomously using the developed Flying Robotic System (FRS) in a commercially produced apple orchard. In this way, it is aimed that producers make accurate yield forecasts before commercial agreements. In this paper, SSD-Mobilenet and Faster R-CNN architecture models trained with COCO datasets referenced in many studies, and SSD-Mobilenet and Faster R-CNN models trained with a learning rate ranging from 0.015-0.04 using the custom dataset are compared experimentally in terms of performance. In the experiments implemented, it is observed that the accuracy rates of the proposed models increased to the level of 93%. Consequently, it has been observed that the Faster R-CNN model, which is developed, makes extremely successful determinations by lowering the loss value below 0.1.

Advancements in deep learning and computer vision have led to the discovery of numerous effective solutions to challenging problems in the field of agricultural automation. With the aim to improve the detection precision in the autonomous harvesting process of green asparagus, in this article, we proposed the DA-Mask RCNN model, which utilizes the depth information in the region proposal network. Firstly, the deep residual network and feature pyramid network were combined to form the backbone network. Secondly, the DA-Mask RCNN model added a depth filter to aid the softmax function in anchor classification. Afterwards, the region proposals were further processed by the detection head unit. The training and test images were mainly acquired from different regions in the basin of the Yangtze River. During the capturing process, various weather and illumination conditions were taken into account, including sunny weather, sunny but overshadowed conditions, cloudy weather, and daytime greenhouse conditions as well as nighttime greenhouse conditions. Performance experiments, comparison experiments, and ablation experiments were carried out using the five constructed datasets to verify the effectiveness of the proposed model. Precision, recall, and F1-score values were applied to evaluate the performances of different approaches. The overall experimental results demonstrate that the balance of the precision and speed of the proposed DA-Mask RCNN model outperform those of existing algorithms.

In underwater environments, ensuring people\'s safety is complicated, with potentially life-threatening outcomes, especially when divers have to work in deeper conditions. To improve the available solutions for working with robots in this kind of environment, we propose the validation of a control strategy for robots when taking objects from the seabed. The control strategy proposed is based on acceleration feedback in the model of the system. Using this model, the reference values for position, velocity and acceleration are estimated, and then the position error signal can be computed. When the desired position is obtained, it is possible to then obtain the position error. The validation was carried out using three different objects: a ball, a bottle, and a plant. The experiment consisted of using this control strategy to take those objects, which the robot carried for a moment to validate the stabilisation control and reference following the control in terms of angle and depth. The robot was operated by a pilot from outside of the pool and was guided using a camera and sonar in a teleoperated way. As an advantage of this control strategy, the model upon which the robot is based is decoupled, allowing control of the robot for each uncoupled plane, this being the main finding of these tests. This demonstrates that the robot can be controlled by a control strategy based on a decoupled model, taking into account the hydrodynamic parameters of the robot.

The real-time detection and counting of rice ears in fields is one of the most important methods for estimating rice yield. The traditional manual counting method has many disadvantages: it is time-consuming, inefficient and subjective. Therefore, the use of computer vision technology can improve the accuracy and efficiency of rice ear counting in the field. The contributions of this article are as follows. (1) This paper establishes a dataset containing 3300 rice ear samples, which represent various complex situations, including variable light and complex backgrounds, overlapping rice and overlapping leaves. The collected images were manually labeled, and a data enhancement method was used to increase the sample size. (2) This paper proposes a method that combines the LC-FCN (localization-based counting fully convolutional neural network) model based on transfer learning with the watershed algorithm for the recognition of dense rice images. The results show that the model is superior to traditional machine learning methods and the single-shot multibox detector (SSD) algorithm for target detection. Moreover, it is currently considered an advanced and innovative rice ear counting model. The mean absolute error (MAE) of the model on the 300-size test set is 2.99. The model can be used to calculate the number of rice ears in the field. In addition, it can provide reliable basic data for rice yield estimation and a rice dataset for research.

Posture changes in pigs during growth are often precursors of disease. Monitoring pigs\' behavioral activities can allow us to detect pathological changes in pigs earlier and identify the factors threatening the health of pigs in advance. Pigs tend to be farmed on a large scale, and manual observation by keepers is time consuming and laborious. Therefore, the use of computers to monitor the growth processes of pigs in real time, and to recognize the duration and frequency of pigs\' postural changes over time, can prevent outbreaks of porcine diseases. The contributions of this article are as follows: (1) The first human-annotated pig-posture-identification dataset in the world was established, including 800 pictures of each of the four pig postures: standing, lying on the stomach, lying on the side, and exploring. (2) When using a deep separable convolutional network to classify pig postures, the accuracy was 92.45%. The results show that the method proposed in this paper achieves adequate pig-posture recognition in a piggery environment and may be suitable for livestock farm applications.

In the process of collaborative operation, the unloading automation of the forage harvester is of great significance to improve harvesting efficiency and reduce labor intensity. However, non-standard transport trucks and unstructured field environments make it extremely difficult to identify and properly position loading containers. In this paper, a global model with three coordinate systems is established to describe a collaborative harvesting system. Then, a method based on depth perception is proposed to dynamically identify and position the truck container, including data preprocessing, point cloud pose transformation based on the singular value decomposition (SVD) algorithm, segmentation and projection of the upper edge, edge lines extraction and corner points positioning based on the Random Sample Consensus (RANSAC) algorithm, and fusion and visualization of results on the depth image. Finally, the effectiveness of the proposed method has been verified by field experiments with different trucks. The results demonstrated that the identification accuracy of the container region is about 90%, and the absolute error of center point positioning is less than 100 mm. The proposed method is robust to containers with different appearances and provided a methodological reference for dynamic identification and positioning of containers in forage harvesting.

Efficiency increase of resources through automation of agriculture requires more information about the production process, as well as process and machinery status. Sensors are necessary for monitoring the status and condition of production by recognizing the surrounding structures such as objects, field structures, natural or artificial markers, and obstacles. Currently, three dimensional (3-D) sensors are economically affordable and technologically advanced to a great extent, so a breakthrough is already possible if enough research projects are commercialized. The aim of this review paper is to investigate the state-of-the-art of 3-D vision systems in agriculture, and the role and value that only 3-D data can have to provide information about environmental structures based on the recent progress in optical 3-D sensors. The structure of this research consists of an overview of the different optical 3-D vision techniques, based on the basic principles. Afterwards, their application in agriculture are reviewed. The main focus lays on vehicle navigation, and crop and animal husbandry. The depth dimension brought by 3-D sensors provides key information that greatly facilitates the implementation of automation and robotics in agriculture.