In recent years, with the continuous innovation of the Internet of Things technology, the image processing technology in the Internet of Things technology has become more and more mature. Automated pig raising will become the mainstream pig raising technology, making the research of image processing technology in the intelligent pig breeding face a change. It has become more and more important. The traditional pig raising model cannot provide a suitable growth and development environment for the pigs. The pigs are disturbed by diseases and environmental discomforts during the growth process, which increases the mortality of the pig breeding process and cannot provide consumers with a guarantee. In order to solve the problem of unfavorable factors during the growth of live pigs, this article uses image processing technology to analyze data through images obtained through automated monitoring and management, uses the system to conduct intelligent, digitized, and standardized management of pig breeding data, and reports to the corresponding. The control module issues instructions to improve the corresponding environmental information and realize the intelligent management of pig breeding. This article will use image processing technology to monitor the growth of pigs in intelligent pig breeding. Studies have shown that the use of image processing technology to realize the intelligent management of pig breeding can help pig farms to carry out manual management to improve production efficiency and management efficiency. The management mode of the pig industry has changed from fuzzy to refined. The breeding cost of pig farms and a lot of manpower and material resources should be reduced, reducing the probability of pigs getting sick and the impact of the environment, reducing the mortality of pigs, improving their economic benefits, and providing consumers with a strong guarantee.

Kinect 3D sensing real-time acquisition algorithm that can meet the requirements of fast, accurate, and real-time acquisition of image information of crop growth laws has become the trend and necessary means of digital agricultural production management. Based on this, this paper uses Kinect real-time image generation technology to try to monitor and study the depth map of crop growth law in real time, use Kinect to obtain the algorithm of crop growth depth map, and conduct investigation and research. Real-time image acquisition research on crop growth trends provides a basis for in-depth understanding of the application of Kinect real-time image generation technology in research. Kinect image real-time acquisition algorithm is a very important information carrier in agricultural information engineering. The research results show that the real-time Kinect depth image acquisition algorithm can obtain good 3D image data information and can provide valuable data basis for the 3D reconstruction of the later crop growth model, growth status analysis, and real-time monitoring of crop diseases. The data shows that, using Kinect, the real-time feedback speed of crop growth observation can be increased by 45, the imaging accuracy is improved by 37, and the related operation steps are simplified by 30. The survey results show that the crop yield can be increased by about 12.

There is an urgent need of developing grape picking robot with intelligent recognition function due to the decrease of grape picking workers’ population. Acquiring the 3D information of picking coordinate is the key process of constructing intelligent picking equipment. In this paper, based on SSD MobileNet neural network model, transfer learning and central deviation angle method were used to realize the positioning of picking coordinate points of facility cultivation grape by machine vision. After testing 720 fruit labels, 633 stem labels and 603 leaf labels labelled by pretreatment, the general precision was 79.5%, which was close to the inherent accuracy of the original model before transfer learning.

In the development of agricultural picking equipment, the identification and positioning of the target fruit is one of the key technologies. There are three sequence steps: finding the target fruit, determining whether the target fruit is mature and needs to be picked, and determining the XYZ coordinate point of the target fruit. In this study, we modified a series of convolution neural networks to detect strawberries on XY scale and judge their maturity. In order to improve the precision of small object detection, we innovatively proposed a new structure called small feature extraction layers. On a moderate GPU, small feature extraction layers improved the positioning precision from 80.53% to 89.96% and small objects precision from 41.58% to 86.14% at a 110 fps object detection speed. When using binocular ranging to measure Z-axis distance, we achieved 15 fps XYZ coordinate determination speed on a 4.5W ultra low-power CPU, and the maximum Z-axis ranging error within 60 cm was less than 2.83%.

This paper presents a valve-controlled pipeline humidification system aimed at achieving precise and uniform humidity regulation in the multi-layer cultivation environments of plant factories. Since grafted seedlings require stable humidity conditions for effective healing, the system was designed to enable fine-grained adjustments across different cultivation layers. A quadratic regression orthogonal rotational combination design was employed to investigate how valve opening angles affect mean relative humidity (MRH), and a regression prediction model was developed accordingly. The model exhibited strong predictive performance, achieving an R² of 0.9907 and an average relative error of only 0.67%. The optimal valve-opening angles were 60°, 50°, and 50°, respectively, which ensured that the MRH remained above 90% throughout operation. Experimental verification confirmed that the model accurately predicted humidity responses, while the proposed system improved uniformity by reducing the humidity variation from 6.1% to 0.3% and increasing the compliance rate from 58.3% to 100%. To enhance short-term humidity forecasting, three machine learning algorithms—Random Forest, XGBoost, and Transformer—were trained to predict humidity trends within a 6-h window. Among them, the RF model achieved the highest accuracy with an R² of 0.9543, outperforming the other models in both stability and precision. The main contribution of this study is the identification of the optimal valve-opening combination through a quadratic orthogonal rotation regression combination experiment. Additionally, the RF obtained the optimal machine learning model for predicting humidity within 6 h.