This project aims to provide growers with the tools needed to adopt site specific weed management (SSWM) strategies as a result of a commercially viable weed ID and mapping system being demonstrated. The adoption of SSWM strategies will result in more efficient use of herbicides, resulting in reduced herbicide usage while providing the desired weed control. This will provide economic benefits to growers through savings on herbicide use and potentially reduce the phytotoxic effect of some herbicides on the crop. It will also deliver environmental and social benefits through reduced herbicide load in cropping systems, resulting in reduced off target impacts on flora and fauna and reduced herbicide residue levels in food. This will benefit the general public and the consumer.
The expected output is a weed ID and mapping system with a demonstrated commercial viability. The benefits of this development will be derived primarily by Australian farmers and applicable to all grain growing regions across the country.
Assessment of the H-Sensor
AgriCon will supply 1 working H-Sensor for this project for testing and development under Australian conditions. The H-Sensor will be a 1 camera system equipped with control unit ready for operation in the field. The H-Sensor will be deployed to Trengove Consulting for testing.
During this assessment, it will be possible to include other potential weed sensing systems for comparison, where SAGIT and other RDC identifies different approaches that require assessment.
The sensor will be deployed to the field for the winter growing season in 2015. Initially the sensors will run with the German based weed classification database and assessed for its ability to detect weeds within the winter crops. The classification database will be modified and fine-tuned to improve the sensor performance with Australian weeds. Images collected from the H-Sensor in the field will be used for further ‘training’ of the sensor. These images will be uploaded to an online software program for classification and inclusion in the database. Access to the online software program is provided by AgriCon.
Using this training technique, existing databases will be improved, and where required new databases will be built to classify new crop types and special interest weed groups. New crop types may include canola, lentils, faba beans, chickpeas and lupins. Weeds of particular interest include annual ryegrass (Lolium rigidum) and wild radish (Raphanus raphanistrum). New and/or refined classification databases will be valuable IP produced by the project. These new databases will be assessed in the field.
The paddocks will be mapped early post emergence, at timings that are in line with herbicide application timings. The H-Sensor will be boom mounted on a ute or quad bike for initial testing. The data from the H-Sensor will be downloaded and maps generated of weed density. The accuracy of the weed maps will be assessed by field observations from specific GPS locations. Field observations of species and species density will be correlated with the classifications generated from the H-Sensor to ascertain its performance in the field.
The initial focus in winter crops will be to;
- discriminate all broadleaf weeds from cereal crops, ensuring that broadleaf weed species of vastly different shapes are correctly classified together using similar weed groups as the GRDC weed ID app, i.e. carrot like (e.g. bifora, fumitory), multiple leaflets (e.g. soursob, medic & clover) and rosettes (e.g. capeweed, sow thistle, indian hedge mustard, wild radish).
- discriminate all grass weeds from broadleaf crops, including canola and the grain legumes lentils, peas, beans, chickpeas and lupins.
Other valuable crop and weed discriminations will be investigated. These include, but are not limited to
- Classifying annual ryegrass in wheat and barley.
- Classifying rosettes (with particular focus on wild radish) from other broadleaf weeds in cereals and legumes.
- Classifying carrot like weeds (with particular focus on bifora) from other broadleaf weeds in cereals and broadleaf crops.
The field testing and training will be conducted in 2015 and 2016. In addition to field testing, weeds of interest will be grown in pots in the glasshouse and imaged using the H-Sensor. These images will be used to assist in building the classification database. This will be conducted out of season, during spring and autumn, providing greater utilisation of the sensor. This would be carried out in the spring/autumn of 2014/15 and 2015/16. Assessing the effect of crop residue on classification accuracy
The ability of the H-Sensor to identify plants emerging through high crop residue levels is a key question to be addressed, as the use of no-till farming techniques in Australia are significantly different to those in Germany. Large field trials will be needed to effectively assess stubble effects. Treatments will include different types and amounts of stubble. This will include treatments with stubble retained standing and stubble removed completely, as well as stubble types including wheat and canola. Field observations will then be made from specific GPS locations within each treatment to compare the effect of crop residue on the accuracy of the H-Sensor.
These trials will be conducted in 2015.