The research on monitoring methods of potato water stress status plays an important role in improving potato growth quality and yield. Based on thermal infrared and visible RGB images, the Crop Water Stress Index (CWSI) can be calculated and used to determine the status of crop water stress. To achieve automated, high-throughput, and non-destructive continuous collection of visible RGB, thermal infrared images, and raw temperature data, this study integrated and developed a crop inspection image acquisition system with binocular cameras and thermal infrared cameras. According to different functional requirements, the hardware parts were designed as different modules, including the main control module, motion control module, image acquisition module, and real-time monitoring module. The main control module implements photo taking control of the image acquisition module, transmission and storage of image and temperature data, and data interaction with the motion control module. Based on the robot operating system (ROS) architecture, multiple nodes were designed to communicate between each module node in publish/subscribe mode, including motion control node, thermal infrared camera node and binocular camera node. Through the one-day system feasibility test in the laboratory, it was concluded that the control accuracy of the motion module meets the needs of image acquisition. Each module and node control program can cooperate to complete image acquisition and local storage. Through the 18-day system stability test and application in the greenhouse, a total of 648 inspections were carried out, each inspection took 3 minutes, and 11664 visible light images, 5832 thermal infrared images and 5832 original temperature data were obtained. It was proved that the system was running stably, the image and temperature data were collected normally, and the function of automatic, high-throughput and non-destructive acquisition of the required data have been realized. The system provides an effective technical and equipment support for crop phenotypic information acquisition.

Navigation of robots has been required to accurately identify the working channel line in the narrow paths inside chicken coops under low lighting conditions. This study aims to detect the centerline of the chicken coop path using 3D LiDAR. Then, 3D LiDAR was equipped on the robot body to collect path information within the operation channel. Various preprocessing techniques were applied, such as direct filtering, ground point filtering, voxel filtering, statistical filtering, and point cloud projection. The point cloud data was obtained to classify in the region of interest (ROI) of the 3D LiDAR on the XOY plane, according to the size of the vertical axis. The center points of the left and right clusters were then selected from the two sets of point cloud datasets after rough classification. The distance between the transverse axis and point was used as the clustering function of K-means clustering to classify the left and right point clouds. Then, the longitude and latitude scanning and the secondary edge methods were used to extract the edges of the two clustered point clouds. The RANSAC was combined to calculate the fitting line equation of the channel edge. The path centerline of the operation channel was extracted using these two equations. An inspection robot was developed for the cage chicken housed to serve as the experimental platform. VLP-16 LiDAR was selected as the perception sensor to conduct the field verification in the D10 and D13 chicken houses of Deqingyuan Co., Ltd. (Beijing, China). The experimental results showed that the improved K-means clustering took an average time of 6.98 ms, with a silhouette coefficient of 0.59, a Rand index of 1.00, a clustering success rate of 84.10%, and a clustering accuracy of 100%. The average time was reduced by 29.40 ms, while the contour coefficient, the rand index, and the accuracy increased by 0.04, 0.63, and 82.41%, respectively, compared with the traditional. The success rate was slightly reduced by 0.41%. The best performance was achieved in both the initial point selection in the clustering function, rather than one single condition. The improved RANSAC shared the accuracy of 93.66% for the centerline extraction and an average error angle of 0.89°, which was 0.14 ° higher than the LSM. The average time (3.94 ms) was reduced by 6.15 ms, compared with the LSM. The improved RANSAC showed a much higher accuracy than before, when the number of iterations was set to 100. Furthermore, the maximum and average absolute error angle were both smaller than before. The improved model can be expected to detect the centerline of chicken coop paths, effectively meeting the actual requirements of real-time autonomous navigation in the cage-style chicken coop environments. The finding can provide the technical support to the autonomous navigation of detection robots in the operation channels of chicken coop.

Crop canopy temperature can often be acquired using the thermal imager. Non-contact and non-destructive automated detection can be expected to achieve for crop water stress status. Automatic image alignment can be used to treat the fuzzy edge distribution, strong noise, as well as shape and texture information lacking in thermal infrared images, according to the information complementarity between visible light and thermal infrared images. The automated extraction can be realized on the crop canopy temperature. This study aims to solve the problems of differences in the radiation, shape, and texture between visible light images and thermal infrared images, leading to the low align images of different modalities. Multimodal image registration was also proposed to integrate the improved brain storm optimization (BSO) and Powell algorithm. Firstly, the original visible light image was downsampled and cropped, according to the normalized cross-correlation value. The area with the most similarity region was obtained in the thermal infrared image under the same resolution; Then, the target area was extracted from the cropped image. The target area image and the original thermal infrared image were decomposed by wavelet transform, where the multilayered low-frequency information was retained; Thirdly, the primitive affine transformation matrix was obtained by the image moments in the low-resolution layer; At the same time, the global search was used to optimize the affine transform matrix in the low-resolution layer using the improved BSO; Fourthly, the optimization was used as the initial point of the Powell algorithm. The optimization was performed in the high-resolution layer; Lastly, the optimization in the previous step was input into the Powell algorithm again. The original image layer was optimized again to obtain the final affine transformation matrix. The original BSO optimization was improved for the optimal affine transformation matrix in the image alignment task. The specific improvements included the following five aspects: The BSO population distribution was initialized using a chaotic mapping function; The mutation range of new individual was modified; The number of K-means clusters was dynamically adjusted in the BSO by the elbow; The chaotic local search was incorporated into the strategy of individual variation; and the probability parameters were dynamically adjusted, according to the different BSO in the early and late stages. Mutual information (MI), normalized mutual information (NMI), root mean square error (RMSE) and mean structure similarity index measure (MSSIM) were taken as the evaluation indexes. A comparison was made with Powell optimization, genetic algorithm (GA) and BSO_Powell algorithm. Specifically, MI indexes were improved by 0.054 2, 0.076 9, 0.040 5, respectively; NMI indexes were improved by 0.015 9, 0.023 1, 0.052 7, respectively; RMSE indexes were reduced by 15.02, 13.03, 27.08, respectively; and MSSIM indexes were improved by 0.052 3, 0.048 8, 0.122 4, respectively, in greenhouse data; In field data, MI indexes were improved by 0.064 2, 0.066 7, 0.035 5, respectively; NMI indexes were improved by 0.007 7, 0.012 5, 0.012 4, respectively; RMSE indexes were reduced by 14.06, 10.57, 15.40, respectively; and MSSIM indexes were improved by 0.047 1, 0.038 1, 0.042 9, respectively. The strong robustness can accurately achieved in the multimodal image registration tasks for potatoes under complex environments.