The wireless sensor network technology provides efficient and reliable technical means for farmland information monitoring in recent years. WSN is a selforganizing network composed of a large number of sensor nodes with sensing and computing capabilities. WSN can detect abnormal events in farmland information, such as fire, environmental pollution, etc. Considering the characteristics of the large monitoring area, limited energy of monitoring nodes and occasional abnormal events, an anomaly event detection for farmland information monitoring based on improved CoSaMP was presented. In the classical CoSaMP algorithm, the choice of similar atom was difficult, and the sparse K required was known. For distinguishing effectively, the correlation between the atoms, the Dice coefficients were used to choose the optimal atom. The PSNR had the similar fluctuation with the match signal residual, which can be used to adjust the number of iterations dynamically. Firstly, the article modeled the farmland sensor network, and optimized the position parameters of the sensor. Then the CoSaMP algorithm was improved, the quality of signal reconstruction was improved by Dice parameters, and the recognition rate of the algorithm was improved by PSNR algorithm. Finally, the algorithm was simulated by Matlab. The simulation results indicated that the algorithms abnormal event detection success rate was 20% higher than that of the existing algorithm, the network energy consumption was reduced by 15%, and the time of detecting was reduced by 50%. At the same time, it provided a theoretical basis for the intelligent monitoring of farmland information and had higher practical application value.