| This research focuses on the problem with common feeding craft in facility pig farm of our country, and aims to achieve real time estimation of pig weight with contactless and automatic means in the process of growth. The key technologies such like dynamic acquisition of pig image, pig dimension detection in real time, pig weight estimation algorithm construction are explored. A new pig weight estimation model is created, and a pig contour extraction algorithm based on depth image is developed. Automatic image filtering algorithm is studied. A mobile pig weight estimation system with contactless and automatic means which is suitable for small group rearing is developed, and it is tested both in lab piggery and real farm. The main conclusions are followed.(1) The pig weight estimation model is studied. Five body dimensions are selected which are propitious both for machine vision detection and manual verify.79 group data is used to build a power regression model based on principal component of body dimensions.97 group data acquired in lab station is used to verify the new model. The correlation coefficient between estimated weight and measured weight is 0.998, the most relative error is less than 4%, and the average relative error is 2.02%.24 group data in real pig farm shows that the average relative error of individual pig is 2.26%±1.78%, and the average error is 2.08kg. The estimation model’s accuracy has been proven.(2) A stereo vision 3D detection system is constructed based on Lab VIEW graphical software development platform. This system is proven to have less than 1% 3D detect relative error within 2m object distance. Both XY axes have a 0.65% average relative detect error, and Z axis’s error is 0.34%. The center of camera has the highest precision in all regions. Pig back 3D measurement and point cloud reconstruction technology are researched; the result is registering with high precision data from 3D laser scanner. There are 255587 points of pig back which have an average error of -3.29±4.51 mm; the 3D detection precision of system is verified.(3) The pig contour extraction algorithm based on depth image is developed, and the problem that the contour extraction algorithm using gray image couldn’t adapt pig harsh light environment has been solved. The relative parameter is used to filter body dimension measurement points for steady extraction. Manual selection function of measurement points is added. Compared with the manual measurement, the measurement result of 32 group pigs in real farm shows that the average relative error of pig body dimension is around 2% and all errors are less than 2cm. A high body detection precision is achieved.(4) The software for pig dimension and weight contactless estimation is developed, and a pig normal posture image filter algorithm using symmetry degree and elongation factor has 79.31% correct selection rate for abnormal image and 73.06 for normal image. The successful detection rate of image captured when pig in free state is 24.93%. Mobile pig weight detection platform is developed to estimate pig weight in multiple piggery which is suitable for small group rearing method in our country. The detected pig capacity is increased and the cost of system is cut off. Three months experiment is carrying out in lab station, the detection precision of body length, body width, body height, hip width and hip height is 1.44%,5.81%,4.94%,2.00% and 1.64% respectively. The average relative error of 268 group data of four pigs is 2.52±2.14%. The pig weight’s automatic and contactless estimation is achieved.(5) TOI technology is used to develop pig product process real-time monitor and manage system, which have data collection function of Environmental and productive process based on C/S framework. Expert decision technology is combined to release information to authorized user through Web station and smart cellphone APP based on B/S framework. The main supervision function includes environment parameter, product information and pig behavior AV surveillance. |
PDF Full Text Request