Traditional methods of cow estrus monitoring technology are not suitable for the current needs of large-scale, intensive and welfare-based farming. There is a need to improve the detection rate of cow estrus and to reduce the emergency response caused by wearing contact devices. Furthermore, it is necessary to verify the practical effectiveness of the LOGISITC and SV (support vector machine) models for modeling cow estrus. In this paper, we have proposed a non-contact cow estrus monitoring method based on the thermal infrared images of cows and have proposed a lab-color-space-based feature extraction method for the thermal infrared images of cow eyes and vulvas. The test subjects were 10 Holstein cows, monitored on a fixed basis, to determine the best segmentation contour. The LOGISTIC and SVM (support vector machine) models were used to establish the cow estrus model using the thermal infrared temperature variation in cows in estrus and cows not in estrus. The experimental results showed that the heat detection rate of the LOGISTIC-based model was 82.37% and the heat detection rate of the SVM-based model was 81.42% under the optimal segmentation profile. The highest temperature in the eye and vulva of cows was the input, and the recall rate was above 86%. The heat monitoring method based on thermal infrared images does not cause stress to cows and meets the needs of modern dairy farming for welfare breeding.

During the crop harvesting process, it is important to obtain the crop yield quickly, accurately and in real time to accelerate the development of smart agriculture. This paper investigated a denoising method applicable to the impact-type sunflower yield sensor signal under the influence of complex noise background in the pneumatic seed delivery structure for a sunflower combine harvester. A signal processing method combining complementary ensemble empirical mode decomposition (CEEMD) and wavelet threshold denoising (WTD) based on an adaptive decomposition capability was proposed by analyzing the non-smoothness of the signal with the impact-type sunflower yield sensor signal in sunflower fields. CEEMD was used to decompose the sunflower seed impact analog signal and field impact-type sunflower yield sensor signal adaptively, and the high frequency components were processed by WTD. Finally the de-noised signal was obtained by reconstruction. An evaluation objective function of the denoising ability of the algorithm based on signal-noise ratio, root mean square error, smoothness and waveform similarity indexes with different weights was also constructed. The results showed that the evaluation objective functions of the simulated and measured signals after denoising by the CEEMD-WTD method are 1.9719 and 4.5318, respectively, which are better than the single denoising methods of EMD (1.5096 and 4.0012), EEMD (1.8248 and 4.0724), CEEMD (1.9516 and 4.3384), and WTD (1.8737 and 4.5294). This method provides a new idea for signal denoising of the impact-type sunflower yield sensor installed in the pneumatic seed delivery structure, and further provides theoretical support and technical references for the development of sunflower high-precision yield measurements in smart agriculture.